1: %%
2: %% rs.tex -- LaTeX2e
3: %%
4: %% Alastair Rucklidge and Mary Silber
5: %% for SIADS
6: %% started 19/10/2005 at INI -- appendix first
7: %% outline 20/01/2006
8: %% further additions June 2007 after Snowbird
9:
10: %% uses bibliography file rs.bib for the references
11:
12: % 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
13: % 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
14: % Digits 0 1 2 3 4 5 6 7 8 9
15: % Exclamation ! Double quote " Hash (number) #
16: % Dollar $ Percent % Ampersand &
17: % Acute accent ' Left paren ( Right paren )
18: % Asterisk * Plus + Comma ,
19: % Minus - Point . Solidus /
20: % Colon : Semicolon ; Less than <
21: % Equals = Greater than > Question mark ?
22: % At @ Left bracket [ Backslash \
23: % Right bracket ] Circumflex ^ Underscore _
24: % Grave accent ` Left brace { Vertical bar |
25: % Right brace } Tilde ~
26:
27: \documentclass{siamltex}
28:
29: \renewcommand{\baselinestretch}{1.2}
30:
31: % packages
32: \renewcommand{\baselinestretch}{1.2}
33: % this is for the mathbb to get Rset below
34: \usepackage{amsfonts}
35: \usepackage{amsmath}
36: \usepackage{xspace}
37:
38: % this collapses lists of numbers 1,2,3,4 to a range 1--4
39: \usepackage{cite}
40:
41: % this is from The Not So Short Introduction to LATEX 2 page 71-77
42: \newif\ifPDF
43: \PDFfalse
44: %\ifx\pdfoutput\undefined\PDFfalse
45: %\else\ifnum\pdfoutput > 0\PDFtrue
46: % \else\PDFfalse
47: % \fi
48: %\fi
49:
50: \ifPDF
51: \usepackage[T1]{fontenc}
52: \usepackage{aeguill}
53: \pdfpagewidth=\paperwidth
54: \pdfpageheight=\paperheight
55: \usepackage[pdftex]{color,graphicx}
56: % \usepackage[pdftex]{graphicx,color}
57: \else
58: % \usepackage[dvips]{graphicx}
59: \usepackage{color,graphicx}
60: \fi
61: % recall: graphicx will use .eps with latex, but .pdf with pdflatex.
62: % to convert: epstopdf <fig>.eps
63:
64: % locally defined commands
65:
66: % \newcommand{\etal}{\hbox{\it et al.}} -- already there
67:
68: % I thought these were defined in the ams packages
69: \newcommand{\Rset}{\mathbb{R}}
70: \newcommand{\Zset}{\mathbb{Z}}
71:
72: % private definitions
73:
74: % surely there is a more elegant way of doing this
75: \def\kdiff{k_{\rm diff}}
76:
77: \def\bfk{\mbox{\mathversion{bold}${k}$}}
78: \def\bfkone{\mbox{\mathversion{bold}${k_1}$}}
79: \def\bfktwo{\mbox{\mathversion{bold}${k_2}$}}
80: \def\bfkthr{\mbox{\mathversion{bold}${k_3}$}}
81: \def\bfkfou{\mbox{\mathversion{bold}${k_4}$}}
82: \def\bfknin{\mbox{\mathversion{bold}${k_9}$}}
83: \def\bfkele{\mbox{\mathversion{bold}${k_{11}}$}}
84: \def\bfktwe{\mbox{\mathversion{bold}${k_{12}}$}}
85:
86: \def\bfm{\mbox{\mathversion{bold}${m}$}}
87: \def\smbfm{\mbox{\mathversion{bold}$\scriptstyle{m}$}}
88: \def\bfkm{{\mbox{\mathversion{bold}${k}$}_{\mbox{\mathversion{bold}$\scriptstyle{m}$}}}}
89:
90: \def\bfx{\mbox{\mathversion{bold}${x}$}}
91:
92: \def\bfsk{\mbox{\mathversion{bold}$\scriptstyle{k}$}}
93: \def\bfskone{\mbox{\mathversion{bold}$\scriptstyle{k_1}$}}
94: \def\bfsktwo{\mbox{\mathversion{bold}$\scriptstyle{k_2}$}}
95: \def\bfskthr{\mbox{\mathversion{bold}$\scriptstyle{k_3}$}}
96:
97: \def\bfsx{\mbox{\mathversion{bold}$\scriptstyle{x}$}}
98:
99: \def\eikdotx{e^{i\bfsk\cdot\bfsx}}
100: \def\eikonedotx{e^{i\bfskone\cdot\bfsx}}
101: \def\eiktwodotx{e^{i\bfsktwo\cdot\bfsx}}
102: \def\eikthrdotx{e^{i\bfskthr\cdot\bfsx}}
103:
104: \def\etwoikonedotx{e^{2i\bfskone\cdot\bfsx}}
105: \def\etwoiktwodotx{e^{2i\bfsktwo\cdot\bfsx}}
106: \def\etwoikthrdotx{e^{2i\bfskthr\cdot\bfsx}}
107:
108: \def\eikonepktwodotx{e^{i(\bfskone+\bfsktwo)\cdot\bfsx}}
109:
110: \def\eikonemktwodotx{e^{i(\bfskone-\bfsktwo)\cdot\bfsx}}
111: \def\eiktwomkthrdotx{e^{i(\bfsktwo-\bfskthr)\cdot\bfsx}}
112: \def\eikthrmkonedotx{e^{i(\bfskthr-\bfskone)\cdot\bfsx}}
113:
114: \def\cc{\mbox{$c.c.$}}
115:
116: \def\semidirectprodsymbol{ %
117: \mathbin{\vrule height1.20ex depth 0.00ex width0.35pt % was 0.25pt
118: \mkern-2.9mu \mathchar"0202}}
119:
120: \def\sdp{%
121: \mskip-\medmuskip \mkern5mu
122: \mathbin{\semidirectprodsymbol}
123: \penalty 900
124: \mkern5mu \mskip-\medmuskip}
125:
126: \title{Design of parametrically forced patterns and quasipatterns}
127:
128: \author{A.M.~Rucklidge\thanks{Department of Applied Mathematics,
129: University of Leeds, Leeds LS2 9JT, UK}
130: \and
131: M.~Silber\thanks{Department of Engineering Sciences and Applied Mathematics,
132: and Northwestern Institute on Complex Systems, Northwestern University,
133: Evanston, IL 60208, USA}}
134:
135: \begin{document}
136:
137: \maketitle
138:
139: \begin{abstract}
140: The Faraday wave experiment is a classic example of a system driven by
141: parametric forcing, and it produces a wide range of complex patterns, including
142: superlattice patterns and quasipatterns. Nonlinear three-wave interactions
143: between driven and weakly damped modes play a key role in determining which
144: patterns are favoured. We use this idea to design single and multi-frequency
145: forcing functions that produce examples of superlattice patterns and
146: quasipatterns in a new model PDE with parametric forcing. We make quantitative
147: comparisons between the predicted patterns and the solutions of the \hbox{PDE}.
148: Unexpectedly, the agreement is good only for parameter values very close to
149: onset. The reason that the range of validity is limited is that the theory
150: requires strong damping of all modes apart from the driven pattern-forming
151: modes. This is in conflict with the requirement for weak damping if three-wave
152: coupling is to influence pattern selection effectively. We distinguish the two
153: different ways that three-wave interactions can be used to stabilise
154: quasipatterns, and present examples of 12-, 14- and 20-fold approximate
155: quasipatterns. We identify which computational domains provide the most
156: accurate approximations to 12-fold quasipatterns, and systematically
157: investigate the Fourier spectra of the most accurate approximations.
158: \end{abstract}
159:
160: \begin{keywords}
161: Pattern formation, quasipatterns, superlattice patterns, mode interactions,
162: Faraday waves.
163: \end{keywords}
164:
165: \begin{AMS}
166: 35B32, 37G40, 52C23, 70K28, 76B15
167: \end{AMS}
168:
169: \pagestyle{myheadings}
170: \thispagestyle{plain}
171: \markboth{A.M. Rucklidge and M. Silber}{Design of parametrically forced
172: patterns and quasipatterns}
173:
174: % ------------------------------------------
175:
176: \section{Introduction}
177: \label{sec:Introduction}
178:
179:
180: The classic Faraday wave experiment consists of a horizontal layer of fluid
181: that spontaneously develops a pattern of standing waves on its surface as it is
182: driven by vertical oscillation with amplitude exceeding a critical value;
183: see~\cite{Arbell2002,Kudrolli1996a,Muller1998a} for recent reviews and surveys.
184: Many other experimental, biological and environmental systems also form
185: patterns~\cite{Cross1993,Hoyle2006}, but Faraday wave experiments have
186: consistently produced patterns with remarkably high degrees of symmetry. One
187: consequence of this is that, over the years, Faraday wave experiments have
188: repeatedly produced new patterns of behaviour that went beyond the existing
189: range of theoretical understanding and required the development of new ideas
190: for their explanation. For example, in the early 1990's, {\em quasipatterns}
191: were discovered in two different Faraday wave experiments, one with a
192: low-viscosity deep layer of fluid with single-frequency
193: forcing~\cite{Christiansen1992,Binks1997,Binks1997b}, and the other with a
194: high-viscosity shallow layer of fluid and forcing with two commensurate
195: temporal frequencies~\cite{Edwards1994}. These patterns are periodic in time
196: but are quasiperiodic in any spatial direction, that is, the amplitude of the
197: pattern (taken along any direction in the plane) can be regarded as the sum of
198: waves with incommensurate spatial frequencies. In spite of this, the spatial
199: Fourier transforms of quasipatterns have 8, 10 or 12-fold rotational order.
200: Quasipatterns are of course related to quasicrystals~\cite{Shechtman1984a}, and
201: quasipatterns have been found in nonlinear optical systems~\cite{Herrero1999},
202: in shaken convection~\cite{Volmar1997,Rogers2005} and in liquid
203: crystals~\cite{Lifshitz2007a} as well as the Faraday wave
204: experiment~\cite{Christiansen1992,Binks1997,Binks1997b,Edwards1994,
205: Kudrolli1998,Arbell2002}. There is, as yet, no satisfactory theoretical
206: understanding of the formation of quasipatterns owing to the problem of small
207: divisors~\cite{Rucklidge2003}.
208:
209: Theoretical efforts aimed at understanding the pattern selection problem have
210: centered around {\em weakly nonlinear theory}. The calculations for the
211: real Faraday wave problem, with finite depth and non-zero viscosity, are
212: difficult (these involve solving the Navier--Stokes equations with a free
213: surface boundary condition~\cite{Skeldon2007}). Most calculations aimed at
214: producing superlattice patterns and quasipatterns have focussed on simpler
215: equations, such as the Zhang--Vi\~nals~\cite{Zhang1996} equations, which model
216: Navier--Stokes when the depth is infinite and the viscosity is small, or on
217: model equations, such as variations on the Swift--Hohenberg
218: equation~\cite{Lifshitz1997,Muller1994,Frisch1995} or the Fitzhugh--Nagumo
219: equations~\cite{Dewel2001}.
220:
221: Nonlinear three-wave resonant interactions have long been recognised as playing
222: a key role in pattern selection in Faraday wave experiments, or other
223: situations where complex patterns are
224: found~\cite{Mermin1985,Newell1993,Edwards1994,Zhang1997}. A series of
225: papers~\cite{Silber1999,Silber2000,Porter2002,Topaz2002,
226: Porter2004a,Topaz2004,Porter2004} has developed this idea, using symmetry
227: considerations to understand pattern selection in Faraday wave experiments with
228: two-frequency forcing, exploiting the three-wave resonant interactions in the
229: context of weakly broken Hamiltonian structure. This approach was able to
230: explain several of the experimentally observed superlattice patterns, and
231: suggested ways of designing multi-frequency forcing functions that could be
232: used to control which patterns would
233: emerge~\cite{Topaz2002,Porter2004a,Topaz2004}. The approach is in principle
234: predictive, but has only been used to determine which additional frequencies to
235: add to the forcing function in order to make observed patterns more
236: robust~\cite{Arbell2002,Ding2006,Epstein2006}. The theory has not been tested
237: quantitatively against solutions of a pattern-forming system, though weakly
238: nonlinear coefficients have been computed for the Zhang--Vi\~nals
239: equations~\cite{Topaz2004,Porter2004} and, more recently, for the
240: Navier--Stokes equations in the infinite depth case~\cite{Skeldon2007}.
241:
242: The main goal of this paper is to come to a greater understanding of some of
243: the complex patterns that are found in high-precision large aspect ratio
244: Faraday wave experiments. Much of the complexity arises from using two or more
245: frequencies in the forcing of the experiment, and recent
246: work~\cite{Porter2004,Topaz2004} explains in principle how to connect the
247: amplitudes and phases of the various components of the forcing frequencies to
248: the nonlinear pattern selection problem. The existing theory provides rules of
249: thumb for designing forcing functions that should encourage the appearance of
250: particular patterns, and it seems to work well, at least qualitatively, and at
251: least in some circumstances~\cite{Arbell2002,Ding2006,Epstein2006}.
252:
253: We take the point of view that in order to claim convincingly that we
254: understand the pattern selection process in these problems with multi-frequency
255: forcing, we should be able to predict {\em in advance} which patterns will be
256: found for different parameter values, we should be able to predict the
257: amplitudes and range of stability of the patterns, and and we should test
258: against a pattern forming system that is not constrained to produce only a
259: limited range of patterns. In order to have a flexible framework for testing
260: predictions, we have devised a partial differential equation (PDE) with
261: multi-frequency forcing~(\ref{eq:pde}) that shares many of the characteristics
262: of the real Faraday wave experiment, but that has easily controllable
263: dissipation and dispersion relations, and simple nonlinear terms. The linear
264: behaviour of the PDE reduces to the damped Mathieu equation, with subharmonic
265: and harmonic tongues, and the simple quadratic and cubic nonlinearities allow
266: three-wave interactions. The PDE has a Hamiltonian limit, but it differs from
267: the real situation in the details of its dispersion relation, and the lack of
268: any coupling to a large-scale mean flow. Notwithstanding these differences, the
269: PDE allows us to explore in detail some of the generic issues surrounding
270: pattern selection in very large aspect pattern forming systems with parametric
271: forcing. A preliminary discussion of the question of quasipattern selection in
272: the new model PDE can be found in~\cite{Rucklidge2007}.
273:
274: In section~\ref{sec:Background}, we review the details of how resonant
275: three-wave interactions influence pattern selection. The main idea is that two
276: pattern-forming modes, with wavevectors separated by an angle~$\theta$, are
277: coupled to a weakly damped mode, and this coupling can lead to the
278: angle~$\theta$ either featuring in the resulting pattern or being eliminated
279: from the resulting pattern~\cite{Mermin1985,Newell1993,Edwards1994,Zhang1997}.
280: We introduce the model PDE in section~\ref{sec:thePDE}, and describe its linear
281: and nonlinear features in sections~\ref{sec:Linear} and~\ref{sec:WNLT}.
282: Appendix~\ref{app:WNLT} gives full details of the weakly nonlinear
283: calculations.
284:
285: In section~\ref{sec:Numericssuperlattice} we devise a forcing function that
286: stabilises the $22^\circ$ superlattice patterns that have been observed in
287: large aspect ratio Faraday wave
288: experiments~\cite{Epstein2006,Kudrolli1998,Arbell2002}. We compute fully
289: nonlinear solutions of the PDE and compare the computed pattern amplitudes with
290: those predicted by weakly nonlinear theory. This demonstrates that the
291: agreement is quantitatively accurate only very close to onset (within~0.1\%).
292: We show that the limit on the range of validity is because of the presence of
293: the weakly damped modes that are required for the superlattice pattern to be
294: stabilised. Thus parameter regimes that are likely to produce the most
295: interesting patterns, arising from coupling to weakly damped modes, are also
296: parameter regimes where weakly nonlinear theory has the most restricted
297: validity.
298:
299: Two mechanisms have been proposed for quasipattern formation, both building on
300: ideas of Newell and Pomeau~\cite{Newell1993}, and one aim of this paper is to
301: demonstrate that both proposed mechanisms for quasipattern formation are viable
302: (preliminary work is reported in~\cite{Rucklidge2007}). One mechanism applies
303: to single frequency forced Faraday waves~\cite{Zhang1996} and has been tested
304: experimentally~\cite{Westra2003}. Another was developed to explain the origin
305: of the two length scales in superlattice patterns~\cite{Topaz2002,Porter2004}
306: found in two-frequency experiments~\cite{Kudrolli1998}. The ideas have not been
307: tested quantitatively, but have been used qualitatively to control
308: quasipattern~\cite{Arbell2002,Ding2006} and superlattice
309: pattern~\cite{Epstein2006} formation in two and three-frequency experiments. We
310: explore the two mechanisms of quasipattern formation in the model PDE in
311: sections~\ref{sec:Numericsquasipatterns}
312: and~\ref{sec:Numericsturbulentcrytals}.
313:
314: Also in section~\ref{sec:Numericsquasipatterns}, we address the distinction
315: between true and approximate quasipatterns, as found in numerical experiments
316: with periodic boundary conditions. Owing to the problem of small divisors,
317: there is as yet no satisfactory mathematical treatment of
318: quasipatterns~\cite{Rucklidge2003}. In spite of this, the weakly nonlinear
319: stability calculations, which are in the framework of a 12-mode amplitude
320: expansion truncated at cubic order, prove to be a reliable guide to finding
321: parameter values where approximate quasipatterns are stable. The fact that
322: stable 12-fold quasipatterns are found where they are expected demonstrates
323: that this approach provides useful information, in spite of the reservations
324: expressed in~\cite{Rucklidge2003}. We explore the effect of domain size on the
325: accuracy of the approximation to a true quasipattern, and show how certain
326: domains yield particularly accurate approximations.
327:
328: In section~\ref{sec:Numericsturbulentcrytals}, we present examples of
329: {\em turbulent crystals}~\cite{Newell1993}: situations in which Fourier modes
330: oriented more than about $20^\circ$ apart do not affect each other, at the
331: level of a cubic truncation. In this context, we find examples of 12-, 14- and
332: 20-fold quasipatterns. These are the first examples of quasipatterns of order
333: greater than 12 found as stable solutions of a~\hbox{PDE} (a preliminary
334: presentation of the 14-fold example is in~\cite{Rucklidge2007}).
335:
336: We summarise our result in section~\ref{sec:Conclusions}.
337:
338: % ------------------------------------------
339:
340: \section{Theoretical background}
341: \label{sec:Background}
342:
343: Resonant triads have played a key role in our understanding of pattern
344: formation~\cite{Mermin1985,Newell1993,Edwards1994,Zhang1997}. This section
345: reviews, somewhat heuristically, the basic selection mechanisms in
346: the case of wave patterns that are parametrically pumped by a two (or more)
347: frequency forcing function. The details behind this summary can be found
348: in~\cite{Porter2004a,Topaz2004}. We write the forcing function as
349: \begin{equation}\label{eq:ft}
350: f(t)=f_m\cos(mt+\phi_m) + f_n\cos(nt+\phi_n) + ...,
351: \end{equation}
352: where $m$ and $n$ are integers, $f_m$ and $f_n$ are real amplitudes,
353: and $\phi_m$ and $\phi_n$ are phases. (We could, of course, set
354: $\phi_m=0$ without loss of generality.) Here we consider $m$~to be
355: the dominant driving frequency, and focus on a pair of waves, each
356: with wavenumber $k_m$, which satisfies the dispersion relation
357: $\Omega(k_m)=m/2$ associated with the linearized problem. In other
358: words, these waves naturally oscillate at a frequency that is
359: subharmonic to the dominant driving frequency $m$, and are typically
360: the easiest to excite parametrically. We write the critical modes in
361: the form $z_1 e^{i\bfskone\cdot\bfsx+imt/2}$ and
362: $z_2 e^{i\bfsktwo\cdot\bfsx+imt/2}$ (together with their complex conjugates),
363: neglecting the higher temporal frequency contributions to the waves.
364: These waves will interact nonlinearly with waves
365: $we^{i\bfskthr\cdot\bfsx+i\Omega(k_3) t}$, where $w$~is a complex amplitude,
366: $\bfkthr=\bfkone+\bfktwo$ and $\Omega(k_3)$ is the frequency
367: associated with~$k_3$, provided that either (1)~the same resonance
368: condition is met with the temporal frequencies, {\em i.e.},
369: $\Omega(k_3) =\frac{m}{2}+\frac{m}{2}$, as in
370: figure~\ref{fig:resonances}(a,b), or (2)~any mismatch
371: $\Delta=|\Omega(k_3)-\frac{m}{2}-\frac{m}{2}|$ in this temporal
372: resonance condition can be compensated for by the forcing
373: function~$f(t)$ (figure~\ref{fig:resonances}c). Case~(1) corresponds
374: to $1:2$ resonance, which occurs even for single frequency forcing
375: ($f_n=0$), and case~(2) applies, for example, to two-frequency forcing
376: with the third wave oscillating at the difference frequency:
377: $\Omega(k_3)=m-n$ and $\Delta=n$. Other cases analogous to~(2), such
378: as $\Omega(k_3)=m+n$, are discussed in~\cite{Topaz2004}, where the
379: special significance of the difference frequency case is explained.
380: Note that in both cases (1) and~(2), the temporal frequency
381: $\Omega(k_3)$ determines the angle $\theta$ between the wave-vectors
382: $\bfkone$ and $\bfktwo$ via the dispersion relation, and therefore
383: provides a possible selection mechanism for certain preferred angles
384: appearing in the power spectrum associated with the
385: wavepatterns. Selecting an angle of $0^\circ$, with $1:2$ resonance in space
386: and time (figure~\ref{fig:resonances}b), is a special case.
387:
388: \begin{figure}
389: \hbox to \hsize{\hfil
390: \hbox to 0.3\hsize{\hfil (a)\hfil}\hfil
391: \hbox to 0.3\hsize{\hfil (b)\hfil}\hfil
392: \hbox to 0.3\hsize{\hfil (c)\hfil}\hfil}
393: \vspace{0.5ex}
394: \hbox to \hsize{\hfil
395: \mbox{\includegraphics[width=0.3\hsize]{two_freq_resonance}}\hfil
396: \mbox{\includegraphics[width=0.3\hsize]{one_freq_resonance_angle_0}}\hfil
397: \mbox{\includegraphics[width=0.3\hsize]{three_freq_resonance}}
398: \hfil}
399: \caption{(a)~$1:2$ resonance occurs between two modes with wavevectors
400: $\bfkone$ and $\bfktwo$ (blue), with the same wavenumber~$k_1$
401: and separated by an angle~$\theta$, and a third mode
402: with wavevector $\bfkthr$ (magenta), provided $\bfkthr=\bfkone+\bfktwo$ and
403: $\Omega(k_3)=2\Omega(k_1)$.
404: (b)~A special case of $1:2$ resonance in space and time occurs for $\theta=0$
405: when the dispersion relation satisfies $\Omega(2k_1)=2\Omega(k_1)$.
406: (c)~With two-frequency $m:n$ forcing, the nonlinear combination of two modes
407: with wavevectors $\bfkone$ and $\bfktwo$, and with $\Omega(k_1)=m/2$ (blue),
408: can, in the presence of a second component of the forcing at frequency~$n$,
409: interact with a mode with wavevector~$\bfkthr$ (red), provided
410: $\bfkthr=\bfkone+\bfktwo$ and $\Omega(k_3)=|m-n|$. Waves driven by frequency~$n$
411: (green) do not enter the resonance condition.}
412: \label{fig:resonances}
413: \end{figure}
414:
415: The nonlinear interactions of the modes can be understood by considering
416: resonant triad equations describing small amplitude standing wave patterns,
417: which take the form
418: \begin{align}\label{eq:resonanttriad}
419: \dot{z}_1&= \lambda z_1+q_1 \bar{z}_2w+(a|z_1|^2+b|z_2|^2)z_1+\cdots\nonumber\\
420: \dot{z}_2&= \lambda z_2+q_1 \bar{z}_1w+(a|z_2|^2+b|z_1|^2)z_2+\cdots\\
421: \dot{w}&= \nu w+q_3z_1z_2+\cdots,\nonumber
422: \end{align}
423: where all coefficients are real, and the dot refers to timescales long compared
424: to the forcing period. Here the quadratic coupling coefficients $q_j$
425: are {\rm O}(1) in the forcing in the $1:2$ resonance case, and {\rm O}$(|f_n|)$
426: in the difference frequency case~\cite{Porter2004}.
427: For other angles $\theta$ between the
428: wavevectors $\bfkone$ and $\bfktwo$ we expect $q_1\approx q_3\approx 0$
429: because the temporal resonance condition for the triad of waves is not met.
430:
431: Since the $m$ frequency is dominant, the $z_1$ and $z_2$~modes will be excited
432: ($\lambda$~increases through zero) while the $w$~mode is damped
433: ($\nu<0$ in~(\ref{eq:resonanttriad})). In this case, $w$~can be
434: eliminated via center manifold reduction~\cite{Guckenheimer1983b} near the bifurcation point
435: ($w\approx \frac{q_3z_1z_2}{|\nu|}$), resulting in the bifurcation
436: problem
437: \begin{align}\label{eq:ampsrhombs}
438: \dot{z}_1&= \lambda z_1+(a|z_1|^2+\tilde{b}|z_2|^2)z_1\nonumber\\
439: \dot{z}_2&= \lambda z_2+(a|z_2|^2+\tilde{b}|z_1|^2)z_2\ .
440: \end{align}
441: These equations describe the competition between standing waves
442: separated by an angle $\theta$, where $\tilde{b}\equiv
443: b+\frac{q_1q_3}{|\nu|}$ explicitly includes the contribution
444: from the slaved mode~$w$, and hence depends on the angle between the two
445: wavevectors $\bfkone$ and~$\bfktwo$.
446:
447: The contribution of the damped~$w$ mode is significant
448: whenever $q_1q_3$ is non-negligible and the damping $|\nu|$ is
449: not too great, and can be made more important by increasing $q_1q_3$
450: and/or by decreasing the damping~$|\nu|$. For instance, in the
451: $1:2$ resonance case, for which $q_1q_3$ is ${\rm O}(1)$, the damping
452: can be decreased by taking $n=2m$ in~(\ref{eq:ft}), since the $w$~mode,
453: with frequency~$m$,
454: will then
455: be driven
456: subharmonically by the $n$~component of the forcing (as well as
457: harmonically by the $m$~component).
458:
459: In the difference
460: frequency case,
461: the quadratic interactions rely on the presence of the $n$~component of the
462: forcing to allow the temporal resonance condition to be met, so
463: $q_1q_3$ is ${\rm O}(|f_n|^2)$. Thus
464: the contribution
465: of the damped~$w$ mode
466: to~$\tilde{b}$
467: can be made more important in two ways:
468: first, by increasing~$f_n$, or second, by parametrically
469: driving the damped mode so that $|\nu|$ is decreased.
470: This requires a third
471: driving frequency $p=2|n-m|$ to be added to the forcing~(\ref{eq:ft}).
472: In both of these
473: instances, the relative phases of the components of the forcing matter,
474: since the
475: modes are being generated nonlinearly with a preferred phase. These
476: ideas are developed systematically in~\cite{Porter2004,Topaz2004},
477: where it is also shown that if there is an underlying Hamiltonian structure, then
478: $q_1q_3<0$ for the $1:2$ resonance, and $q_1q_3>0$ in the
479: difference frequency case provided $n>m$. Note that when $q_1q_3>0$
480: ($q_1q_3<0$) then the $\bfkthr$-mode makes a positive (negative)
481: contribution to the growth rate of the mode $z_2$ when $z_1$ is
482: present, and vice versa.
483:
484: In order to determine more precisely whether the resonant contribution to
485: $\tilde{b}$ is significant enough to lead to a qualitative change in the
486: resulting pattern, it is useful to rescale the amplitudes $z_1$ and $z_2$
487: in~(\ref{eq:ampsrhombs}) by a factor of $1/\sqrt{|a|}$. Then we obtain, for
488: $a<0$, the rescaled equations
489: \begin{align}\label{eq:ampsrhombsnorm}
490: \dot{z}_1&= \lambda z_1-(|z_1|^2+B_{\theta}|z_2|^2)z_1\nonumber\\
491: \dot{z}_2&= \lambda z_2-(|z_2|^2+B_{\theta}|z_1|^2)z_2\ ,
492: \end{align}
493: where $B_{\theta}\equiv \frac{\tilde{b}}{a}$. Here the $\theta$~subscript
494: indicates that the cross-coupling coefficient between the $\bfkone$ and
495: $\bfktwo$ modes depends on the angle~$\theta$ between them.
496:
497: The $B_{\theta}$ function has important consequences for the stability of
498: regular patterns. As a simple example, note that stripes
499: ($|z_1|=\sqrt{\lambda}$, $z_2=0$) are stable if $B_{\theta}>1$, while rhombs
500: associated with a given angle $\theta$
501: ($|z_1|=|z_2|=\sqrt{\lambda/(1+B_{\theta})}$) are preferred if
502: $|B_{\theta}|<1$. Moreover, if $|B_{\theta}|<1$ for any angle $\theta$, then
503: stripes will necessarily be unstable near onset. Since, by judicious choice of
504: forcing frequencies we have at least some ability to control both the magnitude
505: and sign of $B_\theta$ over some range of angles~$\theta$, we have a mechanism
506: for enhancing or suppressing certain combinations of wavevectors in the
507: resulting weakly nonlinear patterns. Alternatively, as suggested
508: by~\cite{Zhang1996}, if we choose forcing frequencies that lead to a large $1:2$
509: resonant contribution at $\theta=0$ (figure~\ref{fig:resonances}b), then the
510: coefficient $a$ can become large compared to $b$, which in turn will cause the
511: rescaled cross-coupling coefficient $B_{\theta}$ to be small over a broad range
512: of angles away from $\theta=0$. (As $\theta\to 0$, it can be shown that
513: $B_\theta\to 2$.) These two cases are contrasted in
514: sections~\ref{sec:Numericsquasipatterns}
515: and~\ref{sec:Numericsturbulentcrytals}, with preliminary work described
516: in~\cite{Rucklidge2007}.
517:
518: Before continuing, we reiterate the constraint on using this analysis to
519: design forcing functions that will stabilise a given pattern. In eliminating
520: the damped mode~$w$, we performed a center manifold reduction, which is valid
521: provided that all non-neutral modes are linearly damped with decay rates
522: that are bounded away from zero. The domain of validity of the reduced
523: equations depends on the extent to which $w$ is damped near the bifurcation
524: point. If that damping is very weak, then the reduced equations will only be
525: quantitatively predictive in a correspondingly small neighborhood of the
526: bifurcation point. We point this out since, from the discussion above, it is
527: clear that the two ways of influencing the magnitude of~$\tilde{b}$, namely
528: increasing the driving force~$f_n$ to control~$q_1q_3$, or driving the
529: difference frequency to reduce the damping~$|\nu|$, both can lead to situations
530: where the center manifold reduction is no longer valid: the $m$ and~$n$ modes
531: could set in together, or the $m$ and difference frequency modes could set in
532: together. In either case, a codimension-two analysis could be performed, but
533: this is beyond the scope of this paper. In practice, the severity of this
534: constraint can only be seen by comparing predictions for the amplitudes and
535: stability of patterns with solutions of the problem at hand (which we do
536: systematically in section~\ref{sec:Numericssuperlattice}).
537:
538: % ------------------------------------------
539:
540: \section{The model PDE}
541: \label{sec:thePDE}
542:
543: In order to explore these issues in detail, we have devised a
544: phenomenological PDE for which the leading nonlinear coefficients in
545: bifurcation problems such as~(\ref{eq:ampsrhombs}), can be calculated
546: relatively easily, and which is reasonably simple to integrate numerically. The
547: equation is:
548: \begin{eqnarray}\label{eq:pde}
549: \frac{\partial U}{\partial t} &=& (\mu+i\omega) U
550: + (\alpha+i\beta)\nabla^2 U
551: + (\gamma+i\delta)\nabla^4U \nonumber\\
552: & & {} + Q_1 U^2 + Q_2 |U|^2 + C |U|^2U
553: + i\hbox{Re}(U) f(t),
554: \end{eqnarray}
555: where
556: $f(t)$ is a real $2\pi$-periodic function, $U(x,y,t)$ is a
557: complex-valued function, with $(x,y)\in\Rset^2$, and $\mu<0$, $\omega$, $\alpha$,
558: $\beta$, $\gamma$ and $\delta$ are real parameters, and $Q_1=Q_{1r}+iQ_{1i}$,
559: $Q_2=Q_{2r}+iQ_{2i}$ and $C=C_{r}+iC_{i}$ are complex parameters.
560:
561: The way the forcing function enters the PDE was chosen so that the linearised
562: problem reduces to the damped Mathieu equation (in much the same way that
563: hydrodynamic models of the Faraday instability reduce to this
564: equation~\cite{Benjamin1954a}). The PDE has the advantage that the dispersion
565: relation can be controlled easily, and weakly nonlinear theory is relatively
566: straightforward to compute. The linear terms are diagonal in Fourier space, so
567: the PDE is readily amenable to the Exponential Time Differencing numerical
568: methods of~\cite{Cox2002}. In addition, the nonlinear terms are simple (they do
569: not involve any derivatives), and so numerical solutions are relatively
570: inexpensive.
571:
572: One of the special features of parametric systems is that, even though pattern
573: selection is a nonlinear process, the position of the linear stability curves
574: determines which resonant triad interactions are dominant. In turn, it is the
575: resonant triad interactions rather than the details of the particular form of
576: nonlinearity in the equation that drives the pattern selection process. For
577: these reasons, the model PDE is a useful testing ground for results derived
578: from symmetric bifurcation theory.
579:
580: The model PDE is similar to the complex Ginzburg--Landau equation -- but we
581: point out that $U(x,y,t)$ is itself the pattern-forming field, and is not the
582: amplitude of some other underlying pattern. With $\mu<0$, all waves are damped
583: in the absence of driving. Note also that the parametric forcing $f(t)$ is
584: explicitly a function of time so that we are resolving dynamics on the fast
585: time-scale set by the periodic forcing. In contrast, other
586: authors~\cite{Coullet1992c,Conway2007,Conway2007a,Halloy2007} have investigated
587: Ginzburg--Landau equations that describe the slow, large spatial-scale
588: evolution of the amplitude of an otherwise spatially homogeneous oscillatory
589: mode arising through Hopf bifurcation. Instead of resolving the fast
590: oscillations of the subharmonic response to the time dependent forcing, a
591: constant-coefficient $\bar U$ term is introduced into the equation,
592: proportional to the amplitude of the parametric forcing~\cite{Coullet1992c}.
593: With multi-frequency forcing, other terms, such as~$\bar U^2$, are also
594: introduced~\cite{Conway2007,Conway2007a,Halloy2007}.
595:
596: There are important qualitative differences, of course, between the model PDE
597: and the Faraday wave experiment. One difference is the role of the $k=0$
598: mode. In the PDE, this mode is damped ($\mu<0$) and has a non-zero
599: frequency~$\omega$; moreover the $k=0$ mode can be nonlinearly excited. In the
600: Faraday wave problem, owing to mass conservation, the $k=0$ mode is neutral and
601: cannot be excited, and this may have important consequences in the
602: dynamics~\cite{Matthews2000,Cox2003}. (In the Zhang-Vi\~nals
603: model~\cite{Zhang1996} this requirement is met since all nonlinear terms
604: appear with an overall spatial derivative that prevents the excitation of the
605: $k=0$ mode.) Another important difference between the model PDE and the Faraday
606: wave experiment is that the dispersion relations have a different structure: in
607: the model PDE, the frequency is a polynomial function of the square of the
608: wavenumber, but the dispersion relation for Faraday waves is more
609: complicated~\cite{Benjamin1954a,Kumar1994,Besson1996}. Nonetheless, the
610: marginal stability curves of the model PDE, especially with multi-frequency
611: forcing, are similar to those that are observed in large aspect ratio Faraday
612: wave experiments.
613:
614: The theory developed by Porter, Topaz and Silber~\cite{Porter2004,Topaz2004}
615: applies in the weakly damped, weakly forced regime, and certain of their
616: results also require that the undamped problem have a Hamiltonian structure.
617: This limit can be realized for our model~(\ref{eq:pde}) by setting
618: $\mu=\alpha=\gamma=C_r=0$ and $Q_2=-2{\bar Q_1}$. In this case, the Hamiltonian
619: is:
620: \begin{align}\label{eq:hamiltonian}
621: H(U,{\bar U}) &= \int\!\!\!\int_R \big[\omega |U|^2 -\beta |\nabla U|^2
622: +\delta |\nabla^2 U|^2+f(t)(\hbox{Re}(U))^2\nonumber\\
623: &\qquad\qquad\qquad\qquad {}-i Q_1 U^2 {\bar U}
624: + i {\bar Q}_1{\bar U}^2 U + \frac{C_i}{2} |U|^4\big]\ dxdy,
625: \end{align}
626: and $U$ evolves according to
627: \begin{equation}
628: \frac{\partial U}{\partial t} = i \frac{\delta H}{\delta{\bar U}}\ .
629: \end{equation}
630: The region $R$ corresponds to the domain of integration of the
631: PDE~(\ref{eq:pde}), where we have assumed periodic boundary conditions
632: apply on $\partial R$. In the examples presented below, some of the parameter
633: choices are nearly Hamiltonian (sections~\ref{sec:Numericsquasipatterns}
634: and~\ref{sec:Numericsturbulentcrytals}) and some are not
635: (section~\ref{sec:Numericssuperlattice}).
636:
637: % ------------------------------------------
638:
639: \section{Linear theory}
640: \label{sec:Linear}
641:
642: The linear problem associated with~(\ref{eq:pde}) takes the form of
643: a damped Mathieu equation for each Fourier mode $\eikdotx$. Specifically, if
644: we set
645: $U(\bfx,t)=\eikdotx (p(t)+iq(t))$ in~(\ref{eq:pde}) linearized about
646: $U=0$, then we obtain
647: \begin{equation}
648: {\ddot p} + {\hat\gamma} {\dot p} + \left(\Omega^2 + {\hat\Omega} f(t)\right) p=0\ ,
649: \end{equation}
650: where
651: \begin{equation}
652: {\hat\gamma}=2\left(-\mu+\alpha k^2-\gamma k^4\right),
653: \qquad
654: {\hat\Omega}=\omega-\beta k^2+\delta k^4,
655: \qquad
656: \Omega^2=\frac{{\hat\gamma}^2}{4} + {\hat\Omega}^2\ .
657: \end{equation}
658: We use the method of~\cite{Besson1996} to solve this linear problem for
659: multi-frequency forcing~$f(t)$, which determines the critical forcing
660: amplitude. Further details can be found in Appendix~\ref{app:WNLT}.
661:
662: We require that the damping $\hat{\gamma}$ be positive for all~$k$, so $\mu<0$,
663: $\gamma\le0$ and $\alpha>-\sqrt{4\gamma\mu}$. However, we do not necessarily
664: insist that it be monotonic, which would require $\alpha\ge0$. Non-monotonic
665: damping is possible in the Faraday wave experiment in shallow layers: if the
666: viscous boundary layers extend from top to bottom of the experimental
667: container, modes with long wavelengths can be more heavily damped than
668: short-wave modes, and indeed the first harmonic mode can be unstable at lower
669: forcing that the subharmonic mode~\cite{Muller1997,Kumar1996a}. We use
670: non-monotonic damping for the example in
671: section~\ref{sec:Numericsquasipatterns}. Likewise, we restrict ourselves to the
672: parameter regime where the dispersion relation $\hat\Omega(k)$ is non-negative
673: and an increasing function of $k^2$; thus we assume $\omega\ge0$, $\delta\ge0$
674: and $\beta\le0$.
675:
676: We will use both $1:2$ and difference frequency resonances to control
677: how modes interact, so it is important to understand that these
678: resonances impose further constraints on the parameters in the
679: dispersion relation. In the case of single frequency forcing
680: $f(t)=a\cos(t+\phi)$, we expect a subharmonic instability to set in
681: first with increasing $a$ at a critical wavenumber $k_1$, which can be
682: estimated by solving the equation
683: ${\hat\Omega}(k_1)=\frac{1}{2}$. This estimate of $k_1$ is good
684: provided the damping is not too large, and that the damping is a
685: (nearly) monotonic function of~$k$. The wavenumber associated with the
686: first harmonic instability is determined by solving
687: ${\hat\Omega}(k_2)=1$. These calculations, together with simple
688: trigonometry, determine that the angle $\theta$ associated with a
689: $1:2$ resonant triad satisfies the equation
690: $k_2=2k_1\cos\left(\frac{\theta}{2}\right)$, which has a solution
691: provided $k_2\le2k_1$. Choosing a scaling of $\bfx$ so that $k_1=1$,
692: then we find that $\omega$, $\beta$ and $\delta$ are related to
693: $\theta$ by
694: \begin{align}
695: {\hat\Omega}(k=1)&= \omega-\beta+\delta=\frac{1}{2}\nonumber\\
696: {\hat\Omega}\left(k=2\cos\left(\frac{\theta}{2}\right)\right)&=
697: \omega-4\beta\cos^2\left(\frac{\theta}{2}\right)
698: +16\delta\cos^4\left(\frac{\theta}{2}\right)=1\ .
699: \end{align}
700: In particular, the $1:2$ resonance will be at $\theta= 0^\circ$ ($k_2=2$)
701: if we choose
702: \begin{equation} \omega= \frac{1}{3}+4\delta\ ,\qquad
703: \beta=-\frac{1}{6}+5\delta\ ,
704: \end{equation}
705: where $\delta\in[0,\frac{1}{30}]$
706: ensures $\beta\le 0$ and hence a monotonic dispersion relation. The $1:2$
707: resonance moves to $\theta= 90^\circ$ ($k_2=\sqrt{2}$) if we choose
708: \begin{equation}
709: \omega= 2\delta\ , \qquad
710: \beta= -\frac{1}{2}+3\delta\ ,
711: \end{equation}
712: where we require $\delta\in[0,\frac{1}{6}]$.
713:
714: Next we consider the case of two-frequency forcing
715: \begin{equation}
716: f(t)=F\left(a_m\cos(mt+\phi_m) + a_n\cos(nt+\phi_n)\right)\ ,
717: \end{equation}
718: where $m$ and $n$ are coprime integers, $(a_m,a_n)$ are relative amplitudes
719: scaled by an overall amplitude~$F$, and $(\phi_m,\phi_n)$ are phases. We
720: focus on a resonant triad involving two critical modes with dominant frequency
721: $\frac{m}{2}$ and a damped difference frequency mode with dominant frequency
722: $n-m$. We will typically take $m$ even and $n$ odd with $n>m$ and
723: $n-m<\frac{m}{2}$ ({\em i.e.}, $\frac{n}{m}\in(1,\frac{3}{2}))$. These
724: conditions imply that the initial instability is expected to be harmonic, that the
725: difference frequency mode decreases $B_\theta$ at the resonance angle in the
726: Hamiltonian limit, and that the difference frequency mode has a wavenumber
727: ($\kdiff$) that is smaller than the critical wavenumber and hence it is not too
728: strongly damped (at least in the case of monotonic dissipation)~\cite{Porter2004}. We estimate
729: the critical wavenumber of instability as $k\approx k_1$, where we assume a
730: scaling such that $k_1=1$. We then have that $k_1$ and $\kdiff$ satisfy
731: \begin{align}
732: {\hat\Omega}(k=1)&=\omega-\beta+\delta=\frac{m}{2}\nonumber\\
733: {\hat\Omega}(k=\kdiff)&=\omega-\beta \kdiff^2+\delta \kdiff^4=n-m\ ,
734: \end{align}
735: which we can solve for $\omega$ and $\beta$:
736: \begin{equation}
737: \omega = \frac{-\kdiff^2\frac{m}{2}+(n-m)}{1-\kdiff^2}
738: + \delta \kdiff^2
739: \quad\hbox{and}\quad
740: \beta = \frac{-\frac{m}{2} + (n-m)}{1-\kdiff^2}
741: + \delta (1+\kdiff^2)\ .
742: \end{equation}
743: Setting $\kdiff=2 \cos\left(\frac{\theta}{2}\right)$, we can
744: relate an angle in the power spectrum of the pattern with the
745: wavenumber $\kdiff$ of the damped mode associated with the
746: resonant triad. For instance, for $\theta=150^\circ$ (the
747: complementary angle to $30^\circ$, which appears in the 12-fold
748: quasipatterns), we have $\kdiff=\sqrt{2-\sqrt{3}}$. Alternatively, if
749: $\theta=158.2^\circ$ (the
750: complementary angle to $21.8^\circ$, which appears in the simplest hexagonal
751: superlattice patterns), we have $\kdiff=\frac{1}{\sqrt{7}}$.
752:
753: % ------------------------------------------
754:
755: \section{Weakly nonlinear theory}
756: \label{sec:WNLT}
757:
758: Our weakly nonlinear calculations are aimed at determining the coefficients of
759: the leading nonlinear terms in finite-dimensional bifurcation problems
760: associated with certain families of patterns in the plane, and the
761: corresponding lattices of wavevectors. These finite-dimensional bifurcation
762: problems allow us to rigorously compute the relative stability of various
763: simple planforms, {\em e.g.}, stripes \hbox{\em vs.} squares, rhombs, hexagons,
764: and also to calculate relative stability of superlattice patterns and hexagons,
765: stripes, and certain rhomb patterns. Moreover these calculations lead to
766: quantitative predictions of the amplitude of the standing wave patterns as a
767: function of the distance $\lambda$ from the bifurcation point, where
768: $\lambda\equiv (F-F_c)/F_c$ and $F_c$ is the critical value of the overall
769: forcing amplitude.
770:
771: A simple example of such a reduction to a finite-dimensional problem was
772: presented in Section~\ref{sec:Background}, where we described the bifurcation
773: problem associated with a pair of standing waves oriented at an angle~$\theta$
774: relative to each other, where $\theta\in(0,\pi/2]$ was bounded away
775: from~$\pi/3$. An example rhombic lattice is shown in
776: figure~\ref{fig:lattices}(a). In that case, the bifurcation problem consisted
777: of a pair of amplitude equations given by~(\ref{eq:ampsrhombsnorm}), and, after
778: appropriate scaling, there was just a single nonlinear coefficient $B_\theta$.
779: The details of the (numerical) computation of this coefficient from the
780: governing PDE~(\ref{eq:pde}) is given in Appendix~\ref{app:WNLT}. In subsequent
781: sections of this paper we present plots of $B_\theta$ as a function of $\theta$
782: for certain parameter sets and forcing functions $f(t)$ used in our numerical
783: simulations of the model PDE.
784:
785: \begin{figure}
786: \hbox to \hsize{\hfil
787: \hbox to 0.24\hsize{\hfil (a)\hfil}\hfil
788: \hbox to 0.24\hsize{\hfil (b)\hfil}\hfil
789: \hbox to 0.24\hsize{\hfil (c)\hfil}\hfil
790: \hbox to 0.24\hsize{\hfil (d)\hfil}\hfil}
791: \vspace{0.5ex}
792: \hbox to \hsize{\hfil
793: \mbox{\includegraphics[width=0.24\hsize]{lattices_rhombs}}\hfil
794: \mbox{\includegraphics[width=0.24\hsize]{lattices_hexagonal_super}}\hfil
795: \mbox{\includegraphics[width=0.24\hsize]{wave_vectors_12_11}}\hfil
796: \mbox{\includegraphics[width=0.24\hsize]{wave_vectors_14_7}}
797: \hfil}
798: \caption{(a)~Rhombic lattice with an angle~$\theta$ between the primary
799: wavevectors. (b)~Hexagonal superlattice, with an angle of $21.8^\circ$ between
800: the most closely spaced wavevectors. (c,d)~12-fold and 14-fold quasilattices,
801: up to 11th order and 7th order respectively~\cite{Rucklidge2003}. See
802: section~\ref{sec:Numericsquasipatterns} for a discussion of how these
803: quasilattices are generated.}
804: \label{fig:lattices}
805: \end{figure}
806:
807: As an additional, specific example we consider patterns associated with the
808: hexagonal superlattice that has been observed in several Faraday wave
809: experiments~\cite{Kudrolli1998,Arbell2002,Epstein2006}. Equivariant bifurcation
810: theory~\cite{Golubitsky1988} was used to derive the form of bifurcation
811: problem~\cite{Dionne1997}. This bifurcation problem describes the long-time
812: evolution of the twelve modes on the critical circle that are associated with
813: patterns that tile a plane in hexagonal fashion. The critical Fourier modes
814: associated with the $21.8^\circ$ superlattice pattern are indicated in
815: Figure~\ref{fig:lattices}(b). We label these modes as follows: $(z_1,z_3,z_5)$
816: are complex amplitudes associated with wavevectors separated by $120^\circ$
817: and, together with their complex conjugates, they comprise the modes associated
818: with one hexagon, while $(z_2,z_4,z_6)$ and their complex conjugates are
819: associated with a second hexagon, rotated by approximately $21.8^\circ$
820: relative to the first.
821:
822: The form of the
823: bifurcation problem associated with this hexagonal superlattice,
824: to cubic order in the amplitudes, is:
825: \begin{align*}
826: \frac{dz_1}{dt} &= \lambda z_1 + Q {\bar z}_3 {\bar z}_5
827: - \left(|z_1|^2 + B_{60}\left(|z_3|^2+|z_5|^2\right)
828: + B_{22}|z_4|^2
829: + B_{38}|z_6|^2
830: + B_{82}|z_2|^2 \right)z_1\\
831: \frac{dz_3}{dt} &= \lambda z_3 + Q {\bar z}_5 {\bar z}_1
832: - \left(|z_3|^2 + B_{60}\left(|z_5|^2+|z_1|^2\right)
833: + B_{22}|z_6|^2
834: + B_{38}|z_2|^2
835: + B_{82}|z_4|^2 \right)z_3\\
836: \frac{dz_5}{dt} &= \lambda z_5 + Q {\bar z}_1 {\bar z}_3
837: - \left(|z_5|^2 + B_{60}\left(|z_1|^2+|z_3|^2\right)
838: + B_{22}|z_2|^2
839: + B_{38}|z_4|^2
840: + B_{82}|z_6|^2 \right)z_5,
841: \end{align*}
842: with similar equations for $z_2$, $z_4$ and $z_6$, related by symmetry. We have shortened
843: the labels of the angles to $22$ instead of $21.8$, {\em etc.} This
844: label indicates the angle between pairs of modes,
845: {\em e.g.}, there is an angle of
846: approximately $82^\circ$ between $z_1$ and $z_2$, while
847: $z_1$ and $z_4$ are separated by approximately $22^\circ$.
848: The
849: nonlinear coefficients $Q$ and $B_{60}$ are computed from the
850: governing PDEs by considering the problem of bifurcation on a
851: simple hexagonal lattice involving a subset of the modes, while
852: $B_{\theta}$ is computed for an
853: arbitrary $\theta\neq60^\circ$ on a rhombic lattice. We can then read
854: off $B_{22}$, $B_{38}$, {\em etc.}, as required.
855: The details are in Appendix~\ref{app:WNLT}.
856: Note that we have assumed that the
857: bifurcation to a stripe planform is supercritical so that we can rescale the
858: amplitudes to make the self--coupling coefficient $a=-1$.
859:
860: The standard planforms, namely stripes, rhombs (associated with each of the angles $22^\circ$,
861: $38^\circ$, $82^\circ$),
862: hexagons, and superlattice patterns, correspond to equilibrium
863: solutions of these equations. The calculation of their linear
864: stability proceeds in a standard fashion and is summarized in~\cite{Dionne1997}.
865: In fact, due to the presence
866: of the quadratic term, all planforms bifurcate unstably, but
867: because $Q$ is typically very small for multi-frequency
868: forcing and sufficiently weak damping~\cite{Porter2002}, the planforms can be stabilized at small
869: amplitude by secondary bifurcations. The
870: superlattice patterns, which satisfy $z_1=z_2=\cdots=z_6$, come in two varieties that bifurcate together
871: and their relative stability is unresolved at cubic order. Specifically, hexagonal
872: superlattice patterns are associated with $z_j$ being real, while
873: triangular superlattice patterns are of the form $z_j=Re^{i\pi/3}$,
874: where $R$ is the real amplitude~\cite{Silber1998}. Which of these two superlattice
875: patterns is favored in a given situation depends on higher order terms
876: in the bifurcation problem. We do
877: not calculate the coefficients of these terms and merely lump the two
878: types of superlattice patterns together in our bifurcation diagrams.
879:
880: % Stripes/hexagons. Here we have:
881: % \begin{align*}
882: % \frac{dz_1}{dt} &= \lambda z_1 + Q {\bar z}_2 {\bar z}_3
883: % - \left(|z_1|^2 + B_{60}|z_2|^2+B_{60}|z_3|^2\right)z_1\\
884: % \frac{dz_2}{dt} &= \lambda z_2 + Q {\bar z}_2 {\bar z}_3
885: % - \left(|z_2|^2 + B_{60}|z_3|^2+B_{60}|z_1|^2\right)z_2\\
886: % \frac{dz_3}{dt} &= \lambda z_3 + Q {\bar z}_2 {\bar z}_3
887: % - \left(|z_3|^2 + B_{60}|z_1|^2+B_{60}|z_2|^2\right)z_3
888: % \end{align*}
889:
890: % quasipatterns.
891: %There are two hexagonal sublattices: $(z_1,z_3,z_5)$ on one, and $(z_2,z_4,z_6)$ on the other, with
892: %$90^\circ$ between $z_1$ and $z_2$. Here we have:
893: % \begin{align*}
894: % \frac{dz_1}{dt} &= \lambda z_1 + Q {\bar z}_3 {\bar z}_5
895: % - \left(|z_1|^2 + B_{60}\left(|z_3|^2+|z_5|^2\right)
896: % + B_{30}\left(|z_4|^2+|z_6|^2\right)
897: % + B_{90}|z_2|^2 \right)z_1\\
898: % \frac{dz_3}{dt} &= \lambda z_3 + Q {\bar z}_5 {\bar z}_1
899: % - \left(|z_3|^2 + B_{60}\left(|z_5|^2+|z_1|^2\right)
900: % + B_{30}\left(|z_6|^2+|z_2|^2\right)
901: % + B_{90}|z_4|^2 \right)z_3\\
902: % \frac{dz_5}{dt} &= \lambda z_5 + Q {\bar z}_1 {\bar z}_3
903: % - \left(|z_5|^2 + B_{60}\left(|z_1|^2+|z_3|^2\right)
904: % + B_{30}\left(|z_2|^2+|z_4|^2\right)
905: % + B_{90}|z_6|^2 \right)z_5
906: % \end{align*}
907: %with similar equations for $z_2$, $z_4$ and $z_6$. This can be obtained from
908: %the superlattice example by renaming $B_{22}$ and $B_{38}$ as~$B_{30}$, and
909: %$B_{82}$ as~$B_{90}$.
910:
911:
912: % \section{Numerical experiments: selecting squares}
913: % \label{sec:Numericssquares}
914: %
915: % For this first example, we will use $1:2$ forcing to design a forcing function
916: % that stabilises square patterns. We will choose parameters so that the mode
917: % driven subharmonically by the $1$~frequency has wavenumber~$k=1$:
918: % $\Omega(k=1)=\frac{1}{2}$. We also want the mode that is driven subharmonically
919: % by the $2$~frequency to have wavenumber~$k=\sqrt{2}$ so as to encourage mode
920: % interactions between modes with wavevectors at $90^\circ$\,:
921: % $\Omega(k=\sqrt{2})=1$. The solutions of this pair of equations are
922: % $\omega=2\delta$ and $\beta=3\delta-\frac{1}{2}$. With the arbitrary choice
923: % $\delta=0.1$, we have $\omega=0.2$ and $\beta=-0.2$. We choose the damping
924: % coefficients in such a way that modes with $k$~away from~1 are strongly damped:
925: % $\mu=-0.2$, $\alpha=-0.2$ and $\gamma=-0.1$. Note that this results in a
926: % damping coefficient that is not a monotonic function of~$k$, but has a minimum
927: % at $k=1$. The linear theory for this problem is shown in
928: % figure~\ref{fig:lineartheorysq}, confirming that with $[1,2]$ forcing, modes
929: % with $k$ close to 1 and $\sqrt{2}$ are neutral and weakly damped respectively.
930: %
931: % % first version had $m=4$, $\delta=0.4$,, $\omega=0.8$, $\alpha=-0.8$,
932: % % $Q_1=0.4(1+i)$, $Q_2=0.8(-1+i)$ and $C=(-1+i)$.
933: %
934: % % first rescale time to get: $m=1$
935: % % $\delta=0.1$, $\omega=0.2$, $\alpha=-0.2$ etc,
936: % % $Q_1=0.1(1+i)$, $Q_2=0.2(-1+i)$ and $C=0.25(-1+i)$.
937: %
938: % % next rescale amplitude by a factor of 2 to get:
939: % % $Q_1=0.2(1+i)$, $Q_2=0.4(-1+i)$ and $C=-1+i$ (since we want $C_r=-1$).
940: %
941: % \begin{figure}
942: % \hbox to \hsize{\hfil
943: % \mbox{\includegraphics[width=0.99\hsize]{rs_fig_squares_linear}}
944: % \hfil}
945: % \caption{Linear theory for the squares example, with $1:2$ resonance in time.
946: % The dispersion relation coefficients are $\omega=0.2$, $\beta=-0.2$ and
947: % $\delta=0.1$, and damping coefficients are $\mu=-0.2$, $\alpha=-0.2$ and
948: % $\gamma=-0.1$.
949: % (a,b) $[1]$~forcing, with $a_1=1$ and $F_c=0.407175$.
950: % (c,d) $[1,2]$~forcing, with $(a_1,a_2)=(1,2.33)$, $(\phi_1,\phi_2)=(0,0)$
951: % and $F_c=0.34867$.
952: % (a,c) show neutral stability curves, whose minima define $F_c$ and the
953: % critical wavenumber (close to 1 in all cases). The curves corresponding to the
954: % primary subharmonic response
955: % are blue, while the harmonic response at twice the basic frequency is magneta.
956: % The corresponding driving frequencies are indicated.
957: % (b,d) show the real parts of the Floquet multipliers.
958: % Note that the primary
959: % instability is subharmonic (the Floquet multiplier is~$-1$) and that in (d),
960: % the wavenumber $k=\sqrt{2}$ is weakly damped (the Floquet multiplier is just
961: % below~$1$).}
962: % \label{fig:lineartheorysq}
963: % \end{figure}
964: %
965: % \begin{figure}
966: % \hbox to \hsize{\hfil
967: % \mbox{\includegraphics[width=0.99\hsize]{rs_fig_squares_btheta}}
968: % \hfil}
969: % \caption{Cross-coupling coefficient $B_{\theta}$ for
970: % $[1]$ (blue) and $[1,2]$ (magenta) forcing,
971: % with linear coefficients as in figure~\ref{fig:lineartheorysq}, and nonlinear
972: % coefficients:
973: % $Q_1=0.2(1+i)$, $Q_2=0.4(-1+i)$ and $C=-1+i$.
974: % The (upper) blue curve is for $[1]$~forcing only, and the (lower) magenta curve
975: % is for $[1,2]$ forcing, showing a strong dip at~$90^\circ$, and giving
976: % $B_{90}=0.25$.}
977: % \label{fig:Bthetasquares}
978: % \end{figure}
979: %
980: % We choose nonlinear coeffients so that the nonlinear part of the PDE is close
981: % to the Hamiltonian limit: $Q_1=0.2(1+i)$, $Q_2=0.4(-1+i)=-2{\bar Q_1}$ and
982: % $C=-1+i$. With forcing at frequency~1 only ($a_1=1$, $a_2=0$), we find a
983: % cross-coupling coefficient that has a peak at~$90^\circ$
984: % (figure~\ref{fig:Bthetasquares}, blue curve). This peak is associated with the
985: % $1:2$ temporal resonance, and occurs at~$90^\circ$ as expected. We can control
986: % the magnitude of this peak (and even turn it into a trough) by including
987: % additional forcing at twice the basic frequency. At this stage, we are aiming
988: % to produce squares, so we need to bring the $B_{\theta}$ curve down
989: % at~$90^\circ$. We achieve this by setting $a_2=2.33$
990: % (figure~\ref{fig:Bthetasquares}, magenta curve), yielding $B_{90}=0.25$.
991: %
992: % As we are close to the Hamlitonian limit, the $1:2$ peak is positive and so
993: % suppresses wavevectors at the corresponding angle~\cite{Porter2002} (
994: % correct reference?). Forcing at frequency~2 will increase or decrease the
995: % height of this peak, depending on the relative phase of the forcing functions
996: % (see below), but an alternative method of controlling the sign of this peak
997: % would be to alter the nonlinear coefficients to move away from the Hamlitonian
998: % limit.
999: %
1000: % \begin{figure}
1001: % \hbox to \hsize{\hfil
1002: % \mbox{\includegraphics[width=0.99\hsize]{rs_fig_squares_b90_theta8}}
1003: % \hfil}
1004: % \caption{Cross-coupling coefficient $B_{90}$ as a function of the
1005: % phase~$\phi_2$ and amplitude~$a_2$ of the second frequency. The other
1006: % parameters are as in figure~\ref{fig:Bthetasquares}. The critical forcing~$F_c$
1007: % also varies with the phase and the amplitude of the second frequency of the
1008: % forcing.}
1009: % \label{fig:B90theta2}
1010: % \end{figure}
1011: %
1012: % In figure~\ref{fig:B90theta2}, we show how the value of the cross-coupling
1013: % coefficient at $90^\circ$, $B_{90}$, depends on the amplitude~$a_2$ and the
1014: % phase~$\phi_2$ of the forcing at frequency~2, for $a_2=0$, $0.5$ and~$1$. The
1015: % magnitude of the variation of~$B_{90}$ increases with~$a_2$, and the smallest
1016: % value of $B_{90}$ is close to $\phi_2=0$. For small~$a_2$, the dependence is
1017: % roughly as $-\cos(\phi_2)$, but for larger~$\phi_2$, there is pronounced
1018: % deviation from the cosine dependence. The reason for this is that the critical
1019: % forcing for the mode associated with~$a_2$ depends on~$\phi_2$, and for
1020: % $\phi_2$ close to $180^\circ$, the parameters are close to the codimension-two
1021: % point where the subharmonic and harmonic modes onset together. For example, at
1022: % $\phi_2=0^\circ$, the codimension-two point occurs when~$a_2=2.4855$, while for
1023: % $\phi_2=180^\circ$, the codimension-two point is at~$a_2=1.4193$. The effect of
1024: % the coupling is inversely proportional to the damping of this mode, which goes
1025: % to zero as we approach the codimension-two point, so the $\phi_2$ dependence is
1026: % strongest close to~$180^\circ$.
1027: %
1028: % dependence appears to look like a cosine. Theory predicts sine, but valid
1029: % only in the limit of small damping. Could check this detail. This is quite
1030: % strong damping. Check vs prediction?
1031: %
1032: % % Aside: the codimension-two points are at:
1033: % % phi_8=0 : $F_c(a_4,a_8)=1.387(1,2.4855)$
1034: % % pih_8-180: $F_c(a_4,a_8)=2.004(1,1.4193)$
1035: %
1036: % \begin{figure}
1037: % \hbox to \hsize{\hfil
1038: % \hbox to 0.31\hsize{\hfil (a)\hfil}\hfil
1039: % \hbox to 0.31\hsize{\hfil (b)\hfil}\hfil
1040: % \hbox to 0.31\hsize{\hfil (c)\hfil}\hfil}
1041: % \vspace{0.5ex}
1042: % \hbox to \hsize{\hfil
1043: % \mbox{\includegraphics[width=0.31\hsize]{rs_fig_squares_stripes}}\hfil
1044: % \mbox{\includegraphics[width=0.31\hsize]{rs_fig_squares_squares}}\hfil
1045: % \mbox{\includegraphics[width=0.31\hsize]{rs_fig_squares_squares_16_seed2}}\hfil}
1046: % \caption{With a forcing~$F=1.1F_c$, we find (a)~stripes and (b,c)~squares with
1047: % (a)~single frequency forcing, and (b,c)~two-frequency forcing, as in
1048: % figure~\ref{fig:Bthetasquares}. In~(a,b), The square domain was four critical
1049: % wavelengths in each direction, with $128^2$ Fourier modes. In~(c),
1050: % squares are found in a domain sixteen times the critical wavelength. The
1051: % grey-scale represents the real part of $U(x,y,t)$ with $t$ equal to an integer
1052: % multiple of~$2\pi$.}
1053: % \label{fig:stripessquares}
1054: % \end{figure}
1055: %
1056: % Since we are looking to decrease $B_{90}$ below~1, we choose $\phi_2=0$ (close
1057: % to the minima of the curves), and increase~$a_2$ to~$2.33$, close to the
1058: % codimension-two point to have a strong enough effect
1059: % (figures~\ref{fig:lineartheorysq} and~\ref{fig:Bthetasquares}).
1060: % Figure~\ref{fig:stripessquares} confirms this qualitative success of this
1061: % choice: at $1.1$ times critical, stripes are stable with single frequency
1062: % forcing, while our choice of~$a_2=2.33$ is enough to stabilize squares with
1063: % two-frequency forcing in $4\times4$ and $16\times16$ domains. In fact, with
1064: % random initial conditions in the larger domain, several different modulated
1065: % square-like patterns can be stable, but the difference between these and true
1066: % squares is beyond the range of this theory.
1067: %
1068: % \begin{figure}
1069: % \hbox to \hsize{\hfil
1070: % \mbox{\includegraphics[width=0.99\hsize]{rs_fig_squares_bifn}}
1071: % \hfil}
1072: % \caption{Amplitudes of stripes and squares as a function of $\lambda$ for
1073: % (a)~single frequency forcing, and (b) two-frequency forcing, as in
1074: % figure~\ref{fig:Bthetasquares}. Using weakly nonlinear theory, we denote stable
1075: % solutions with solid lines and unstable solutions with dashed lines. We also
1076: % denote numerically computed solutions of the PDEs by $+$~for stripes (in black)
1077: % and $\Box$~for squares (in magenta). In~(b), the PDE stripe solutions are
1078: % unstable. (c)~is a detail of~(b) with additional data points.}
1079: % \label{fig:stripesvssquares}
1080: % \end{figure}
1081: %
1082: % In figure~\ref{fig:stripesvssquares}, we make a quantitative comparison between
1083: % the amplitudes of stripes and squares as numerical solutions of the PDE in a
1084: % domain big enough for one repeat of the patter, and the values predicted by
1085: % weakly nonlinear theory, for forcing up to 1.1~times critical, and for single
1086: % and two-frequency forcing. In both cases, the numerical solutions of the PDEs
1087: % agree with the weakly nonlinear prediction, provided we are close enough to
1088: % onset. For stripes, the agreement is quantitative for the whole of the range
1089: % (up to 1.1 time critical and beyond), but for squares, there is quantitative
1090: % agreement only up to about 1.002 times critical
1091: % (figure~\ref{fig:stripesvssquares}c). We have confirmed the quantitative
1092: % agreement very close to onset by plotting the data on a logarthmic scale. The
1093: % reason for this lack of quantitative agreement for squares for only a modestly
1094: % large degree of supercriticality is discussed in section~\ref{sec:Discussion}.
1095:
1096: % ------------------------------------------
1097:
1098: \section{Numerical experiments: selecting superlattice patterns}
1099: \label{sec:Numericssuperlattice}
1100:
1101: Motivated by experimental
1102: observations~\cite{Epstein2006,Kudrolli1998,Arbell2002} of superlattice
1103: patterns in the Faraday wave experiment with $6:7$ forcing, in this section
1104: we use $6:7$ forcing, with some additional forcing at frequency~2 to drive
1105: the difference frequency mode, in order to stabilise a $21.8^\circ$ hexagonal
1106: superlattice pattern. We carry out the linear and weakly nonlinear calculations
1107: to find parameter values for which hexagonal superlattice patterns are
1108: predicted to be stable, and solve the model PDE numerically to confirm these
1109: predictions. In addition, we make quantitative comparisons between the weakly
1110: nonlinear predictions and the numerical solutions of the PDEs, comparing the
1111: predicted amplitudes and ranges of stability of the patterns. The agreement is
1112: not quantitative except at very small amplitude, and we develop an explanation
1113: for this at the end of the section.
1114:
1115: The PDE was solved numerically using the fourth-order Runge--Kutta
1116: Exponential Time Differencing numerical method (ETD4RK)
1117: of~\cite{Cox2002}. This pseudospectral method solves the linear part exactly,
1118: excluding the parametric forcing term, which is included with the
1119: nonlinear terms. This allows the use of a timestep based on
1120: accuracy requirements, rather than numerical stability
1121: limits. Timestepping takes place in spectral space, and the
1122: timestep is chosen to be one-twentieth of the shortest of the periods
1123: of the forcing function~$f(t)$, in order that the effect of the
1124: time-dependent forcing is fully resolved by the fourth-order
1125: method. The nonlinear terms are evaluated using
1126: FFTW~\cite{Frigo2005}. The resolution was relatively low for the
1127: examples in this section (up to $96\times56$ Fourier modes),
1128: but we used up to $1536^2$ Fourier modes
1129: for the largest quasipattern examples discussed below. At each
1130: timestep, the upper half of the Fourier spectrum was removed in order
1131: to dealias the cubic terms.
1132:
1133: \begin{figure}
1134: \hbox to \hsize{\hfil
1135: \mbox{\includegraphics[width=0.99\hsize]{rs_fig_superlattice_linear}}
1136: \hfil}
1137: \caption{An example of the linear theory for one, two and three-frequency
1138: forcing, with dispersion relation coefficients $\omega=2/3$, $\beta=-7/3$ and
1139: $\delta=0$, and damping coefficients $\mu=-0.1$, $\alpha=0.01$ and
1140: $\gamma=-0.1$.
1141: (a,b) $6$~forcing, with $a_6=1$ and $F_c=0.83973$.
1142: (c,d) $6:7$~forcing, with $(a_6,a_7)=(1,1)$, $(\phi_6,\phi_7)=(0,0)$
1143: and $F_c=0.80839$.
1144: (e,f) $6:7:2$~forcing, with $(a_6,a_7,a_2)=(1,1,0.45)$,
1145: $(\phi_6,\phi_7,\phi_2)=(0,0,240^\circ)$, $F_c=0.80975$ and $k_c=0.9910$.
1146: (a,c,e) show neutral stability curves, whose minima define $F_c$ and the
1147: critical wavenumber (close to 1 in all cases). Curves corresponding to the
1148: response to frequency~6 are blue, to frequency~7 are green, and to twice the
1149: difference frequency are red, with the corresponding driving frequency
1150: indicated. The minimum of the red curve in (e) is close to $k=\frac{1}{\sqrt{7}}$.
1151: (b,d,e) show the real parts of the Floquet multipliers at the critical
1152: forcing.}
1153: \label{fig:lineartheorysl}
1154: \end{figure}
1155:
1156: \begin{figure}
1157: \hbox to \hsize{\hfil
1158: \mbox{\includegraphics[width=0.99\hsize]{rs_fig_superlattice_btheta}}
1159: \hfil}
1160: \caption{$B_{\theta}$, with linear parameters as in
1161: figure~\ref{fig:lineartheorysl} and nonlinear coefficients $Q_1=2+i$,
1162: $Q_2=1+2i$, $C=-1+30i$. The blue curve is with $6$~forcing only:
1163: the dip at $90^\circ$ is because of $1:2$ resonance with frequency~6.
1164: The green curve has $6:7$ forcing: note the dip starting at $22^\circ$,
1165: corresponding to the difference frequency, even though this frequency is not
1166: forced directly. The red curve has $6:7:2$ forcing: note the pronounced dip
1167: at $22^\circ$, and the smaller dip around $43^\circ$, corresponding to
1168: frequency~4 (which is in $1:2$ resonance with frequency~$2$).
1169: $B_{60}$~is calculated separately, as described in Appendix~\ref{app:WNLT}.
1170: The relevant coefficients are $B_{22}=0.46$, $B_{38}=1.23$, $B_{60}=1.11$ and
1171: $B_{82}=0.63$.}
1172: \label{fig:Bthetasuperlattice}
1173: \end{figure}
1174:
1175: \begin{figure}
1176: \hbox to \hsize{\hfil
1177: \hbox to 0.3\hsize{\hfil (a)\hfil}\hfil
1178: \hbox to 0.3\hsize{\hfil (b)\hfil}\hfil
1179: \hbox to 0.3\hsize{\hfil (c)\hfil}\hfil}
1180: \vspace{0.5ex}
1181: \hbox to \hsize{\hfil
1182: \mbox{\includegraphics[width=0.3\hsize]{rs_fig_superlattice_hexagons}}\hfil
1183: \mbox{\includegraphics[width=0.3\hsize]{rs_fig_superlattice_superlattice}}\hfil
1184: \mbox{\includegraphics[width=0.3\hsize]{rs_fig_superlattice_rect22}}\hfil}
1185: \caption{With $6:7:2$ frequency forcing at (a,b)~1.004 and (c)~1.02
1186: times critical, we find hexagons, a superlattice pattern, and
1187: $22^\circ$~rectangles. The resolution was $96\times56$ Fourier modes, and
1188: the domain is rectangular, $2\sqrt{7}\times2\sqrt{7/3}$ critical
1189: wavelengths, big enough to fit two copies of the superlattice pattern.
1190: The grey-scale represents the real part of $U(x,y,t)$ with $t$ equal to an
1191: integer multiple of~$2\pi$.}
1192: \label{fig:hexsuperrect}
1193: \end{figure}
1194:
1195: \begin{figure}
1196: \hbox to \hsize{\hfil
1197: \mbox{\includegraphics[width=0.99\hsize]{rs_fig_superlattice_bifn}}
1198: \hfil}
1199: \caption{Amplitudes of hexagons (blue), the superlattice pattern (red),
1200: $22^\circ$ rectangles (magenta) and stripes (black) as a function of $\lambda$
1201: for $6:7:2$ forcing. Using weakly nonlinear theory, we denote stable
1202: solutions with solid lines and unstable solutions with dashed and dotted lines.
1203: We also denote numerically computed stable solutions of the PDEs by $+$~for
1204: hexagons, $\times$ for the superlattice pattern, $\Box$~for rectangles, and
1205: $\diamond$~for (unstable) stripes.
1206: The amplitudes of the PDE solutions are computed by matching the time evolution
1207: of the $k=1$ modes to the linear response functions.
1208: (b)~is a detail of~(a), showing that the agreement between weakly nonlinear theory
1209: and the PDEs improves very close to onset.}
1210: \label{fig:bifurcationsuperlattice}
1211: \end{figure}
1212:
1213: We choose parameters so that the mode driven subharmonically by the
1214: $6$~frequency has wavenumber~$k=1$: ${\hat\Omega}(k=1)=3$. The wavenumber associated
1215: with $21.8^\circ$ is $\frac{1}{\sqrt{7}}$, which we wish to correspond to the difference
1216: frequency ($7-6$), so we set ${\hat\Omega}(\frac{1}{\sqrt{7}})=1$. This mode responds
1217: subharmonically to driving at frequency~2. Furthermore, the
1218: wavenumber associated with $81.8^\circ$ is part of this superlattice pattern, and
1219: we can influence this if we set ${\hat\Omega}(\frac{4}{\sqrt{7}})=6$, in $1:2$ resonance with
1220: the primary response. This yields $\omega=\frac{2}{3}$, $\beta=-\frac{7}{3}$
1221: and $\delta=0$. We choose damping coefficients $\mu=-0.1$, $\alpha=0.01$ and
1222: $\gamma=-0.15$, and nonlinear coefficients $Q_1=2+i$, $Q_2=1+2i$ and
1223: $C=-1+30i$. Note that these parameters are not close to the Hamiltonian limit:
1224: in fact, we have chosen the nonlinear coefficients so that the $1:2$
1225: interaction {\em reduces} the cross-coupling coefficient in the range of angles close
1226: to~$90^\circ$.
1227:
1228: The linear theory for this problem is shown in figure~\ref{fig:lineartheorysl},
1229: confirming that modes with $k$ close to 1 and $\frac{1}{\sqrt{7}}$ are neutral
1230: and weakly damped respectively, with $6:7:2$ forcing
1231: (figure~\ref{fig:lineartheorysl}f). For this example, we have set the phases of
1232: the two main components of the forcing equal to zero, and (after some
1233: experimentation) set the phase of the component that drives the difference mode
1234: equal to~$240^\circ$.
1235:
1236: The cross-coupling coefficient~$B_{\theta}$ is shown in
1237: figure~\ref{fig:Bthetasuperlattice}: with $6$~forcing only, there is a dip in
1238: the curve around $90^\circ$ owing to $1:2$ resonance
1239: at~$82^\circ$. (Had we chosen parameters sufficiently close to a Hamiltonian limit, this feature would have
1240: been a peak rather than a dip~\cite{Porter2004}.)
1241: With $6:7$~forcing, the dip at $22^\circ$, corresponding to
1242: the difference frequency, is visible, even though this frequency is not forced
1243: directly. Finally, with $6:7:2$~forcing, we can control the depth of the dip
1244: at~$22^\circ$. An additional feature at $43^\circ$ is visible, corresponding to
1245: frequency~4, in $1:2$ resonance with frequency~$2$.
1246:
1247: Solutions of the PDE with $6:7:2$~forcing are shown in
1248: figure~\ref{fig:hexsuperrect}, confirming that hexagons, superhexagons and
1249: $22^\circ$ rectangles are all stable solutions for different parameter values.
1250:
1251: In figure~\ref{fig:bifurcationsuperlattice}, we make quantitative comparison
1252: between the amplitudes and stability of these patterns as numerical solutions
1253: of the PDE, and the values predicted by weakly nonlinear theory, for forcing up
1254: to 1.02~times critical. In addition, we show the results of one-dimensional
1255: simulations, which recover the unstable stripe pattern. In all cases,
1256: numerical solutions of the PDE agree with the weakly nonlinear prediction,
1257: provided we are close enough to onset (we have confirmed this by plotting the
1258: data on a logarithmic scale). Secondary bifurcations, which delimit parameter
1259: intervals where the patterns are stable, are also recovered, although the
1260: agreement is only qualitatively correct.
1261:
1262: It is notable that the agreement between the amplitudes predicted by weakly
1263: nonlinear theory and measured from PDE simulations is not particularly good for
1264: the multi-mode patterns, when compared to the much better agreement in the case of
1265: stripes. The reason for this lack of quantitative agreement for the complex
1266: patterns can be understood by going to higher order in the center manifold reduction that was
1267: performed to go from (\ref{eq:resonanttriad}) to~(\ref{eq:ampsrhombs}). We
1268: recall the framework: there are two weakly excited modes with amplitudes~$z_1$
1269: and $z_2$, and a damped mode~$w$, which evolve according to:
1270: \begin{align}
1271: \dot{z}_1&= \lambda z_1+q_1 \bar{z}_2w+(a|z_1|^2+b|z_2|^2)z_1,\nonumber\\
1272: \dot{z}_2&= \lambda z_2+q_1 \bar{z}_1w+(a|z_2|^2+b|z_1|^2)z_2,\\
1273: \dot{w}&= \nu w+q_3z_1z_2,\nonumber
1274: \end{align}
1275: where all coefficients are real, and $\nu<0$. We have discarded those
1276: higher order terms that do not play a role in the centre manifold
1277: reduction in order to emphasise the effect of the higher-order nonlinear terms
1278: that appear as a result of the reduction. We express
1279: $w$ on the centre manifold as a power series in $z_1$ and~$z_2$, and
1280: perform a centre manifold reduction to find
1281: \begin{equation}
1282: w = -\frac{q_3}{\nu}z_1z_2
1283: + \frac{q_3(q_1q_3 -(a+b)\nu)}{\nu^3}\left(|z_1|^2+|z_2|^2\right)z_1z_2 + \cdots,
1284: \end{equation}
1285: which results in
1286: \begin{align}
1287: \dot{z}_1&= \lambda z_1+(a|z_1|^2+\tilde{b}|z_2|^2)z_1
1288: + \frac{q_1q_3(q_1q_3-(a+b)\nu)}{\nu^3}\left(|z_1|^2+|z_2|^2\right)|z_2|^2z_1+\cdots,
1289: \nonumber\\
1290: \dot{z}_2&= \lambda z_2+(a|z_2|^2+\tilde{b}|z_1|^2)z_2
1291: + \frac{q_1q_3(q_1q_3-(a+b)\nu)}{\nu^3}\left(|z_1|^2+|z_2|^2\right)|z_1|^2z_2+\cdots,
1292: \label{eq:amplitudesquintic}
1293: \end{align}
1294: where $\tilde{b}=b-q_1q_3/\nu$ as before. In order for the $w$~mode to
1295: influence the coupling constant $B_\theta=\tilde{b}/a$, and so produce
1296: interesting patterns, the quadratic coefficients $q_1$ and $q_3$ must be
1297: non-zero (the modes must be in three-wave resonance) and $\nu$ must be
1298: small (the mode must be weakly damped). However, if $\nu$ is small, the
1299: $\nu^3$ in the denominator of the quintic terms imply that these
1300: high-order terms will be important exactly where the most interesting patterns
1301: will be found. Indeed, the graphs of amplitude against~$\lambda$ in
1302: figure~\ref{fig:bifurcationsuperlattice} are
1303: well fit by a quintic polynomial.
1304:
1305: In contrast, the stripe pattern involves damped modes at $k=0$ and $k=2$, and
1306: it can be seen from figure~\ref{fig:lineartheorysl}(f) that these modes are
1307: well damped, so there is no reason for quantitative agreement not to
1308: extend to larger values of~$\lambda$, and indeed it does.
1309:
1310: One can estimate the range of validity of the cubic truncation
1311: of~(\ref{eq:amplitudesquintic}) when $\lambda$ and $\nu$ are both small. In the
1312: case of rhombs, the linear and cubic terms balance when
1313: $|z_1|^2=|z_2|^2={\mathcal O}(\lambda\nu)$. The quintic terms are thus smaller than the
1314: linear and cubic terms when $\lambda\ll\nu$, which is what one would expect:
1315: the center manifold reduction is valid when all modes that are eliminated are
1316: heavily damped compared to the modes that are retained.
1317:
1318: Therefore, this codimension-one approach to finding interesting patterns has
1319: the smallest range of validity exactly where the patterns are most likely to be
1320: interesting. A proper treatment would require consideration of the
1321: codimension-two problem $(\lambda,\nu)=(0,0)$, which is beyond the
1322: scope of this paper.
1323:
1324: A further complication in the Faraday wave situation (and in our model PDE) is
1325: that in order for the interaction between the primary harmonic modes (driven by
1326: the $m$~forcing) and the
1327: weakly damped difference frequency modes ($n-m$) to take place, the subharmonic
1328: mode~($n$) must be present in the forcing function: the quadratic
1329: coefficients~$q_1$ and~$q_3$ increase with the strength of the subharmonic
1330: forcing~$f_n$, so the interaction is strongest when $f_n$ is largest. This
1331: implies that the subharmonic mode is itself only weakly damped, and will be
1332: excited if~$f_n$ is increased beyond its critical value. Experimental evidence
1333: suggests that the codimension-two point (or bicritical point), where the
1334: primary harmonic ($m$) and subharmonic ($n$) modes are both neutral, is an
1335: organising centre for the dynamics~\cite{Edwards1994,Kudrolli1998,Arbell2002}.
1336: (There has been relatively little progress on the theoretical understanding of
1337: the bicritical point, apart from a study in the case of a single
1338: frequency~\cite{Wagner2003} and in a few particular cases for two-frequency
1339: excitation~\cite{Porter2002,Porter2004a}.) Therefore, the problem should really
1340: be treated as a codimension-three interaction between primary harmonic modes
1341: and subharmonic modes, as well as weakly damped harmonic modes.
1342:
1343: Notwithstanding these complications, it is clear that the idea that
1344: pattern selection is being influenced by three-wave coupling to weakly damped
1345: modes is fundamentally correct, and the codimension-one approach, while having
1346: limited quantitative agreement with PDE simulations, is clearly providing the
1347: correct qualitative interpretation of the observed patterns.
1348:
1349: % ------------------------------------------
1350:
1351: \section{Numerical experiments: 12-fold quasipatterns}
1352: \label{sec:Numericsquasipatterns}
1353:
1354: In the previous section, we demonstrated how to stabilise simple patterns by
1355: driving the difference frequency. In this section, we show how this mechanism
1356: can be used to predict parameter values for stable approximate 12-fold
1357: quasipatterns, and we demonstrate how well a periodic pattern in a large domain
1358: can approximate a quasipattern.
1359:
1360: \begin{figure}
1361: \hbox to \hsize{\hfil
1362: \mbox{\includegraphics[width=0.99\hsize]{rs_fig_qp_linear}}
1363: \hfil}
1364: \caption{Linear theory for the quasipattern example. The dispersion relation
1365: coefficients are $\omega=0.633975$, $\beta=-1.366025$ and $\delta=0$, and the
1366: damping coefficients are $\mu=-0.2$, $\alpha=-0.2$ and $\gamma=-0.15$.
1367: (a,b) $4$~forcing, with $a_4=0.57358$ and $F_c=1.04730$.
1368: (c,d) $4:8$~forcing, with $(a_4,a_8)=(0.57358,1.6)$, $(\phi_4,\phi_8)=(0,0)$
1369: and $F_c=0.95214$.
1370: (e,f) $4:5:8$~forcing, with $(a_4,a_5,a_8)=(0.57358,0.81915,1.6)$,
1371: $(\phi_4,\phi_5,\phi_8)=(0,0,0)$, $F_c=0.93159$ and $k_c=0.9798$.
1372: (a,c,e) show neutral stability curves, whose minima define $F_c$ and the
1373: critical wavenumber (close to 1 in all cases). Harmonic curves are blue and
1374: subharmonic curves are green, with the corresponding driving frequency
1375: indicated.
1376: (b,d,e) show the real parts of the Floquet multipliers.}
1377: \label{fig:lineartheoryqp}
1378: \end{figure}
1379:
1380: In order to use triad interactions to encourage modes at $30^\circ$, we choose
1381: $m=4$, $n=5$ forcing: $4:5$~forcing has been used in several experiments to
1382: produce 12-fold quasipatterns~\cite{Kudrolli1998,Edwards1994,Arbell2000}. We
1383: set ${\hat\Omega}(k=1)=2$ so that the subharmonic response to frequency~4 comes
1384: at wavenumber~1, and we require that a wavenumber involved in $150^\circ$ mode
1385: interactions ($k^2=2-\sqrt{3}$) correspond to the difference frequency:
1386: ${\hat\Omega}(k)=1$. One solution is $\omega=0.633975$, $\beta=-1.366025$ and
1387: $\delta=0$. Twelve-fold quasipatterns also require modes at $90^\circ$ to be
1388: favoured, and for this choice of parameters, ${\hat\Omega}(k=\sqrt{2})$
1389: is~$3.37$. Although this is not particularly close to~$4$, we can use $1:2$
1390: resonance (driving at frequency~8) to control the $90^\circ$ interaction. The
1391: linear theory for these cases is in figure~\ref{fig:lineartheoryqp}, with
1392: $\mu=-0.2$, $\alpha=-0.2$ and $\gamma=-0.15$. Note that the damping is
1393: non-monotonic, and has a minimum at $k=1$.
1394:
1395: \begin{figure}
1396: \hbox to \hsize{\hfil
1397: \mbox{\includegraphics[width=0.99\hsize]{rs_fig_qp_btheta}}
1398: \hfil}
1399: \caption{Cross-coupling coefficient $B_{\theta}$ for the parameters from
1400: figure~\ref{fig:lineartheoryqp} and nonlinear coefficients $Q_1=1+i$,
1401: $Q_2=-2+2i$, $C=-1+10i$. The blue curve is with $4$ forcing only: the peak
1402: near $80^\circ$ is because of $1:2$ resonance with frequency~8. The magenta
1403: curve has $4:8$ forcing, using the frequency~8 component to bring down the
1404: curve close to~$90^\circ$. Bringing
1405: in the 5~frequency (red curve) give a pronounced dip at $30^\circ$,
1406: corresponding to the difference frequency~$(5-4)$, even though this mode is not
1407: driven directly. The relevant coefficients are $B_{30}=-0.01$, $B_{60}=2.24$
1408: and $B_{90}=-0.51$.}
1409: \label{fig:BthetaQP}
1410: \end{figure}
1411:
1412: \begin{figure}
1413: \hbox to \hsize{\hfil
1414: \mbox{\includegraphics[width=0.99\hsize]{rs_fig_qp_bifn}}
1415: \hfil}
1416: \caption{Bifurcation diagram with 3-frequency $4:5:8$ forcing based on a
1417: 12-amplitude cubic truncation, with parameter values as in
1418: figure~\ref{fig:BthetaQP}. 12-fold quasipatterns (red) are predicted to be
1419: stable up to $1.0085$ times critical; squares (magenta) are stable from
1420: $1.0013$ times critical and $30^\circ$~rectangles (green) from $1.0092$ times
1421: critical. The crosses show the amplitudes of stable approximate quasipattern
1422: solutions of the PDE, calculated in a domain $2\sqrt{13}\times2\sqrt{13/3}$
1423: critical wavelengths.}
1424: \label{fig:QPbifn}
1425: \end{figure}
1426:
1427: The resulting $B_{\theta}$ curve with $4:5:8$ forcing and with
1428: near-Hamiltonian choice of nonlinear coefficients (figure~\ref{fig:BthetaQP})
1429: shows pronounced dips at $30^\circ$ and $90^\circ$ as required. In
1430: figure~\ref{fig:QPbifn}, we show that within a 12-amplitude cubic truncation,
1431: 12-fold quasipatterns are stable between $0.9995$ and $1.0085$ times critical.
1432: We have found stable approximate quasipatterns (marked by crosses on the
1433: figure, and discussed in more detail below) in the same range. Squares are also
1434: stable above $1.0013$ times critical. The agreement is good, better than the
1435: examples in the previous section (figure~\ref{fig:bifurcationsuperlattice}),
1436: since the difference frequency mode is still fairly well damped (compare the
1437: damping for $k$ close to $0.378$ in figure~\ref{fig:lineartheorysl}f and
1438: $k=0.518$ in figure~\ref{fig:lineartheoryqp}f).
1439:
1440: \subsection{Choice of domain size for approximate quasipatterns}
1441:
1442: Before presenting numerical solutions in large domains,
1443: we discuss the issue of choice of domain for providing accurate
1444: approximations to quasipatterns.
1445:
1446: Just as hexagonal patterns can be approximated in rectangular
1447: domains~\cite{Matthews1998}, there are many ways of choosing periodic domains
1448: to allow accurate approximations of quasipatterns. Here we discuss three
1449: plausible approaches to choosing domains for 12-fold quasipatterns. We show why
1450: one approach, based on Pythagorean triplets, does not work at all, while two
1451: other approaches both work well.
1452:
1453: Reducible symmetry group representations with square periodic domains have
1454: been put forward as candidates for producing approximate
1455: quasipatterns~\cite{Solari1997,Dawes2003}. In particular, Pythagorean triplets
1456: have been identified as of particular interest in this case~\cite{Dawes2003}.
1457: Consider a pair of integers~$(p,q)$ with $p>q>0$. Then $(p^2-q^2,2pq,p^2+q^2)$
1458: forms a Pythagorean triplet ({\em i.e.}, $(p^2-q^2)^2+(2pq)^2=(p^2+q^2)^2)$,
1459: and the vectors $\bfkone=(1,0)$ and $\bfktwo=(2pq,p^2-q^2)/(p^2+q^2)$ have the
1460: same length. For example, with $p=7$, $q=4$, we have $\bfkone=(1,0)$ and
1461: $\bfktwo=(\frac{56}{65},\frac{33}{65})$. If $\frac{p}{q}$ is a continued
1462: fraction approximation of~$\sqrt{3}$ (table~\ref{table:sqrtthree}), then the
1463: angle between these two vectors, namely
1464: $\tan^{-1}\left((p^2-q^2)/(2pq)\right)$, tends to $30^\circ+\mathcal{O}(1/q^2)$
1465: as the approximation to~$\sqrt{3}$ improves. Thus it might be thought that
1466: square $(p^2+q^2)\times(p^2+q^2)$ domains might readily allow approximations to
1467: 12-fold quasipatterns.
1468:
1469: Unfortunately, these Pythagorean domains do not provide good approximations to
1470: quasipatterns. The reason is that the essential $60^\circ$~coupling is not
1471: quite correct: consider the wavevector $\bfkone=(p^2+q^2,0)$ (in units of the
1472: basic lattice vector), and, at $60^\circ$ on either side of~$\bfkone$ there are
1473: the wavevectors $\bfkthr=(p^2-q^2,2pq)$ and $\bfkele=(p^2-q^2,-2pq)$. In order
1474: for quadratic interactions between these three modes to occur, we need
1475: $\bfkone=\bfkthr+\bfkele$. However, in this case,
1476: $\bfkthr+\bfkele=(2(p^2-q^2),0)=\bfkone+(p^2-3q^2,0)$, which is close to, but
1477: never equal to,~$\bfkone$: it can be shown that $p^2-3q^2=1$ or~$-2$, which is
1478: small is compared to $|\bfkone|=p^2+q^2$. This small difference means that the
1479: important nonlinear interactions between these three waves generate erroneous
1480: long-wave modulations in square $(p^2+q^2)\times(p^2+q^2)$ domains, and if
1481: modes with wavenumber close to zero are not heavily damped, these long-wave
1482: modulations can dominate the pattern.
1483:
1484: \begin{table}
1485: \begin{center}
1486: \begin{tabular}{|c|l|}
1487: \hline $\strut$
1488: $\frac{p}{q}\strut$ & $\frac{2}{1}$, $\frac{5}{3}$,
1489: $\frac{7}{4}$, $\frac{19}{11}$, $\frac{26}{15}$, $\frac{71}{41}$,
1490: $\frac{97}{56}$, $\frac{265}{153}$, $\frac{362}{209}$,
1491: $\cdots{}\rightarrow\sqrt{3}$\\
1492: \hline
1493: \end{tabular}
1494: \vspace{2mm}
1495: \caption{Continued fraction approximations to $\sqrt{3}$.}
1496: \label{table:sqrtthree}
1497: \end{center}
1498: \end{table}
1499:
1500: \begin{table}
1501: \begin{center}
1502: \begin{tabular}{|c|l|l|l|l|}
1503: \hline $\strut$
1504: $p/q$ & $A=\mbox{Area}$ & First two wavevectors &
1505: $(|\bfktwo|-1)A$ &
1506: $(\angle_{12}-30^\circ)A$\\
1507: \hline $\strut$
1508: \strut$\frac{p}{q}$ & $\frac{4(p^2-3pq+3q^2)}{\sqrt{3}}$
1509: & $\bfkone=\frac{((p-q)\sqrt{3},3q-p)}{2\sqrt{p^2-3pq+3q^2}}$,
1510: &
1511: & \\
1512: \strut &
1513: & $\bfktwo=\frac{((2q-p)\sqrt{3},p)}{2\sqrt{p^2-3pq+3q^2}}$
1514: &
1515: & \\
1516: \strut$\frac{7}{4}$ & $30.02$
1517: & $\frac{\left(3\sqrt{3},5\right)}{\sqrt{52}}$,
1518: $\frac{\left(\sqrt{3}, 7\right)}{\sqrt{52}}$
1519: & $\phantom{-}0$
1520: & $\phantom{-}66.2$\\
1521: \strut$\frac{19}{11}$ & $224.01$
1522: & $\frac{\left(8\sqrt{3},14\right)}{\sqrt{338}}$,
1523: $\frac{\left(3\sqrt{3},19\right)}{\sqrt{338}}$
1524: & $\phantom{-}0$
1525: & $-132.3$\\
1526: \strut$\frac{26}{15}$ & $418.00$
1527: & $\frac{\left(11\sqrt{3},19\right)}{\sqrt{724}}$,
1528: $\frac{\left(4 \sqrt{3},26\right)}{\sqrt{724}}$
1529: & $\phantom{-}0$
1530: & $\phantom{-}66.2$\\
1531: \strut$\frac{71}{41}$ & $3120.0$
1532: & $\frac{\left(30\sqrt{3},52\right)}{\sqrt{5404}}$,
1533: $\frac{\left(11\sqrt{3},71\right)}{\sqrt{5404}}$
1534: & $\phantom{-}0$
1535: & $-132.3$\\
1536: \strut$\frac{97}{56}$ & $5822.0$
1537: & $\frac{\left(41\sqrt{3},71\right)}{\sqrt{10084}}$,
1538: $\frac{\left(15\sqrt{3},97\right)}{\sqrt{10084}}$
1539: & $\phantom{-}0$
1540: & $\phantom{-}66.2$\\
1541: \strut\vdots & & & & \\
1542: \strut$\sqrt{3}$ & $\infty$
1543: & $\frac{\left(1,1\right)}{\sqrt{2}}$,
1544: $\frac{\left(\sqrt{3}-1,\sqrt{3}+1\right)}{2\sqrt{2}}$
1545: & $\phantom{-}$ & $\phantom{-}$ \\
1546: \hline $\strut$
1547: \strut$\frac{p}{q}$ & $2q\times2q$
1548: & $\bfkone=(1,0)$,
1549: $\bfktwo=\left(\frac{p}{2q},\frac{1}{2}\right)$
1550: &
1551: & \\
1552: \strut$\frac{7}{4}$ & 64
1553: & $\bfkone=(1,0)$,
1554: $\bfktwo=\left(\frac{7}{8},\frac{1}{2}\right)$
1555: & $\phantom{-}0.498$
1556: & $-16.3$ \\
1557: \strut$\frac{19}{11}$ & 484
1558: & $\bfkone=(1,0)$,
1559: $\bfktwo=\left(\frac{19}{22},\frac{1}{2}\right)$
1560: & $-1.001$
1561: & $\phantom{-}33.2$ \\
1562: \strut$\frac{26}{15}$ & 900
1563: & $\bfkone=(1,0)$,
1564: $\bfktwo=\left(\frac{26}{30},\frac{1}{2}\right)$
1565: & $\phantom{-}0.500$
1566: & $-16.5$ \\
1567: \strut$\frac{71}{41}$ & 6724
1568: & $\bfkone=(1,0)$,
1569: $\bfktwo=\left(\frac{71}{82},\frac{1}{2}\right)$
1570: & $-1.000$
1571: & $\phantom{-}33.1$ \\
1572: \strut$\frac{97}{56}$ & 12544
1573: & $\bfkone=(1,0)$,
1574: $\bfktwo=\left(\frac{97}{112},\frac{1}{2}\right)$
1575: & $\phantom{-}0.500$
1576: & $-16.5$ \\
1577: \strut\vdots & & & & \\
1578: \strut$\sqrt{3}$ & $\infty$
1579: & $\bfkone=(1,0)$,
1580: $\bfktwo=\left(\frac{\sqrt{3}}{2},\frac{1}{2}\right)$
1581: & $\phantom{-}$ & $\phantom{-}$ \\
1582: \hline
1583: \end{tabular}
1584: \vspace{2mm}
1585: \caption{Domains that provide good approximations to 12-fold quasipatterns.
1586: The first column gives the rational number $\frac{p}{q}$ that is a continued
1587: fraction approximation to~$\sqrt{3}$, drawn from table~\ref{table:sqrtthree}.
1588: The second and third columns give the area of a computational domain and
1589: two of the
1590: wavevectors that will make up an approximate quasipattern. The fourth and fifth
1591: columns demonstrate that the errors in the length of $\bfktwo$ and in the
1592: angle~$\angle_{12}$
1593: between $\bfkone$ and $\bfktwo$ scale as $1/A$, or equivalently, as~$q^{-2}$.
1594: The first set of rows refer to rectangular domains of size
1595: $2\sqrt{p^2-3pq+3q^2}\times2\sqrt{(p^2-3pq+3q^2)/3}$, which allow superlattice
1596: patterns that approximate quasipatterns. In these examples, all wavevectors
1597: are the same length. The second set of rows refer to square
1598: domains of size $2q\times2q$. In these domains, approximate quasipatterns have
1599: wavevectors that have two slightly different lengths.
1600: }
1601: \label{table:domains}
1602: \end{center}
1603: \end{table}
1604:
1605: A much better way of generating good approximations to 12-fold quasipatterns is
1606: to choose $2q\times2q$ domains, with vectors $\bfkone=(1,0)$ and
1607: $\bfktwo=(p,q)/2q$, with $\frac{p}{q}$ drawn from table~\ref{table:sqrtthree}.
1608: Again, the angle between these vectors goes as $30^\circ+\mathcal{O}(1/q^2)$,
1609: and, though the wavenumbers are not quite equal, we have
1610: $|\bfktwo|\rightarrow1+\mathcal{O}(1/q^2)$. This approach works because with
1611: this choice of wavevectors, we do have the correct three-wave coupling:
1612: $\bfkthr=(q,p)/2q$, $\bfkele=(q,-p)/2q$ and so $\bfkthr+\bfkele=\bfkone$.
1613:
1614: A third possibility is to approximate the quasipattern as a superlattice
1615: pattern, using the 12-dimensional irreducible representations of~$D_6\sdp
1616: T^2$~\cite{Dionne1997}. If we choose $\alpha=q$ and $\beta=p-q$ (in the
1617: notation of~\cite[table~2]{Dionne1997}), with $\frac{p}{q}$ drawn from
1618: table~\ref{table:sqrtthree}, then
1619: $\bfkone=((p-q)\sqrt{3},3q-p)$,
1620: $\bfktwo=((2q-p)\sqrt{3},p)$,
1621: $\bfkthr=((p-2q)\sqrt{3},p)$,
1622: $\bfkele=((q\sqrt{3},3q-2p)$ (all these should be divided
1623: by their length, $\sqrt{p^2-3pq+3q^2}$). We have
1624: $\bfkthr+\bfkele=\bfkone$ as required, all wavevectors are the same length
1625: (which is an advantage over the second alternative), and the angle between
1626: $\bfkone$ and $\bfktwo$ goes as $30^\circ+\mathcal{O}(1/q^2)$. One
1627: disadvantage of this approach is that numerical solutions of the PDE must be
1628: carried out in $2\sqrt{p^2-3pq+3q^2}\times2\sqrt{(p^2-3pq+3q^2)/3}$ rectangular
1629: domains in order to take advantage of spectral numerical methods. These domains
1630: are big enough to contain two repeats of the pattern (as in
1631: figure~\ref{fig:hexsuperrect}b) and so only half the Fourier coefficients are
1632: used.
1633:
1634: The last two methods are compared in table~\ref{table:domains}. The error
1635: between the approximation and the 12-fold quasipattern is inversely
1636: proportional to the area of the domain in both cases. The hexagonal
1637: superlattice approximations have all wavevectors of the correct length, unlike
1638: the square approximation. For this reason, the superlattice case is better for
1639: computing bifurcation diagrams, as all modes bifurcate at the same value of the
1640: forcing.
1641:
1642: However, for similar domain areas, the wavevectors in the square case are about
1643: four times closer to $30^\circ$ apart than the superlattice case. Moreover, the
1644: square case is more amenable to efficient use of fast Fourier transforms, since
1645: the number of modes can be chosen to be a power of two in each direction, while
1646: the superlattice case leads to more awkward choices of numbers of modes.
1647:
1648: For these reasons, we choose the rectangular $2\sqrt{13}\times2\sqrt{13/3}$
1649: domain for the bifurcation diagram in figure~\ref{fig:QPbifn}, since the
1650: quantitative comparison between numerical simulation and analysis is easier
1651: when all modes have the same wavenumber and hence bifurcate at the same value
1652: of the forcing (see below). We use the more convenient square ($8\times8$,
1653: $30\times30$ and $112\times112$) domains for the remaining PDE simulations
1654: described below. It would be interesting to see how sensitive the quasipattern
1655: is to the exact choice of domain size. It would also be interesting to find
1656: domains that provide particularly accurate approximations to 8-, 10-, 14-fold
1657: and higher order quasipatterns, but these issues are beyond the scope of this
1658: paper.
1659:
1660: One consequence of the wavevectors being of unequal length in the square case
1661: is that the two wavenumbers concerned have slightly different critical
1662: forcings. In the $30\times30$ and $112\times112$ cases, this difference is
1663: negligible, but in the $8\times8$ case, the critical forcings for the
1664: two wavenumber are appreciably different. We therefore make a small adjustment
1665: to the domain size to make the two critical forcings the same, raised by a
1666: factor of $1.00066$ above~$F_c$.
1667:
1668: \begin{figure}
1669: \hbox to \hsize{\hfil
1670: \hbox to 0.45\hsize{\hfil (a)\hfil}\hfil
1671: \hbox to 0.45\hsize{\hfil (b)\hfil}\hfil}
1672: \vspace{0.5ex}
1673: \hbox to \hsize{\hfil
1674: \mbox{\includegraphics[width=0.45\hsize]{rs_fig_qp8}}\hfil
1675: \mbox{\includegraphics[width=0.45\hsize]{rs_fig_qp30}}\hfil}
1676: \vspace{1ex}
1677: \hbox to \hsize{\hfil (c)\hfil}
1678: \vspace{0.5ex}
1679: \hbox to \hsize{\hfil
1680: % \mbox{\includegraphics[width=0.95\hsize]{rs_fig_qp112_23}}\hfil}
1681: \mbox{\includegraphics[width=0.90\hsize]{rs_fig_qp112_6030}}\hfil}
1682: \caption{Numerical solutions of the PDE with $4:5:8$ forcing at $1.003$ times
1683: critical.
1684: (a) $8\times8$ domain.
1685: (b) $30\times30$ domain.
1686: (c) $112\times112$ domain (only a $60\times30$ section of the domain is
1687: shown).
1688: Parameter values are as in figure~\ref{fig:BthetaQP}.}
1689: \label{fig:QPexamples}
1690: \end{figure}
1691:
1692: \subsection{Numerical examples of 12-fold quasipatterns}
1693:
1694: Numerical solutions of the PDE~(\ref{eq:pde}) with $4:5:8$ forcing at $1.003$
1695: times critical are shown in figure~\ref{fig:QPexamples}, in periodic domains
1696: $8\times8$, $30\times30$ and $112\times112$ wavelengths with periodic boundary
1697: conditions and using up to $1536^2$ Fourier modes. The solutions were followed
1698: for at least $10\,000$ forcing periods in the largest domain. Most initial
1699: conditions resulted in square patterns, but minor adjustments to the Fourier
1700: amplitudes at an early stage of the calculation resulted in stable approximate
1701: 12-fold quasipatterns. Note, however, that the PDE solutions in
1702: figure~\ref{fig:QPexamples} were not constrained to chose exactly the
1703: wavevectors given in table~\ref{table:domains}.
1704:
1705: The accuracy of the approximation improves with increasing domain size. The
1706: modes involved in the $30\times30$ example are $(30,0)$ and $(26,15)$ and their
1707: reflections, which are $29.98^\circ$ apart, and differ in length by $0.05\%$.
1708: The larger $112\times112$ domain allows an improved approximation to the
1709: quasipattern: the important wavevectors are $(112,0)$ and $(97,56)$, which are
1710: $29.9987^\circ$ apart and differ in length by $0.004\%$. The amplitudes of
1711: these modes differ by~$1\%$. We discuss other ways of evaluating the improved
1712: approximation to quasiperiodicity in the next section.
1713:
1714: \subsection{Fourier spectra of quasipatterns}
1715:
1716: All the numerical PDE solutions presented here have been carried out in
1717: periodic domains, so these solutions are only approximations to quasipatterns.
1718: In the 12-fold examples (figures~\ref{fig:QPexamples} and \ref{fig:NPexamples}a
1719: below), the most important twelve wavevectors in the pattern are not exactly
1720: $30^\circ$ apart and do not have exactly the same length (see
1721: table~\ref{table:domains}). On the other hand, it
1722: is possible to generate true quasipatterns using twelve modes with $k=1$ evenly
1723: spaced around the unit circle, but the asymptotic series in the weakly
1724: nonlinear approximation are known to diverge~\cite{Rucklidge2003}. In this
1725: section, we compare the detailed Fourier spectra of the approximate
1726: quasipatterns to see how these differ from the spectra of true quasipatterns.
1727:
1728: We make the comparison by computing the locations of modes generated by
1729: nonlinear interactions up to a certain order, in the three cases of approximate
1730: quasipatterns in figure~\ref{fig:QPexamples}, as well as in a true
1731: quasipattern. To do this, we first define the {\em order} of a mode.
1732:
1733: Quadratic interactions between the twelve modes with wavevectors $\bfkone$,
1734: \dots, $\bfktwe$ generate new modes with
1735: wavevectors $2\bfkone$, $\bfkone+\bfktwo$, $\bfkone+\bfkthr$ and so on.
1736: Nonlinear interactions of~$N$ of the twelve modes
1737: generate modes with wavevectors~$\bfkm$:
1738: \begin{equation}
1739: \bfkm=\sum_{j=1}^{12} m_j \bfk_j,
1740: \end{equation}
1741: where the $m_j$'s are non-negative integers adding up to~$N$. We define
1742: $|\bfm|=\sum_{j=1}^{12}m_j$. In the case of a periodic domain, the set of all
1743: possible~$\bfkm$ defines a lattice of wavevectors
1744: (figure~\ref{fig:lattices}a,b), while in the quasipattern case, the set of all
1745: possible~$\bfkm$ defines a {\em quasilattice}: examples of 12- and 14-fold
1746: quasilattices with $|\bfm|\leq11$ ($|\bfm|\leq7$ in the 14-fold case) are shown
1747: in figure~\ref{fig:lattices}(c,d).
1748:
1749: If a wavevector~$\bfk$ is in the lattice or quasilattice, then $\bfk=\bfkm$ for
1750: an infinite number of choices of vector~$\bfm$, since increasing $m_1$ and
1751: $m_7$ (say) by the same amount does not change~$\bfkm$ but increases $|m|$
1752: by~2. We define the {\em order} of wavevector $\bfk$ to be the smallest value
1753: of~$|\bfm|$ with $\bfkm=\bfk$.
1754:
1755: \begin{figure}
1756: \hbox to \hsize{\hfil(a)\hfil}
1757: \vspace{0.5ex}
1758: \hbox to \hsize{\hfil
1759: \mbox{\includegraphics[width=0.8\hsize]{rs_fig_qp_spectra_07}}\hfil}
1760: \vspace{1ex}
1761: \hbox to \hsize{\hfil(b)\hfil}
1762: \vspace{0.5ex}
1763: \hbox to \hsize{\hfil
1764: \mbox{\includegraphics[width=0.8\hsize]{rs_fig_qp_spectra_11}}\hfil}
1765: \vspace{1ex}
1766: \hbox to \hsize{\hfil(c)\hfil}
1767: \vspace{0.5ex}
1768: \hbox to \hsize{\hfil
1769: \mbox{\includegraphics[width=0.8\hsize]{rs_fig_qp_spectra_15}}\hfil}
1770: \caption{Fourier spectra of $8\times8$ (cyan), $30\times30$ (blue),
1771: $112\times112$ (red) approximate 12-fold quasipatterns, as well as the 12-fold
1772: quasilattice (small black dots), up to (a)~7th, (b)~11th and (c)~15th order.
1773: The size of the coloured marker indicates the amplitude of the corresponding
1774: Fourier mode on a logarithmic scale, with the smallest markers having
1775: $10^{17}$ times smaller amplitude than the largest. Only Fourier modes close to
1776: the unit circle ($0.8\leq|\bfk|\leq1.2$) are shown, with wavevector angles
1777: $-10^\circ\leq\theta\leq40^\circ$. The horizontal line $k=1$ represents the
1778: unit circle, while the curved line represents $k_x=1$. Wavevectors that come
1779: closest to the unit circle up to a particular order are labelled
1780: with~$\times$.}
1781: \label{fig:QPFourierSpectra}
1782: \end{figure}
1783:
1784: \begin{figure}
1785: \hbox to \hsize{\hfil
1786: \hbox to 0.45\hsize{\hfil (a)\hfil}\hfil
1787: \hbox to 0.45\hsize{\hfil (b)\hfil}\hfil}
1788: \vspace{0.5ex}
1789: \hbox to \hsize{\hfil
1790: \mbox{\includegraphics[width=0.45\hsize]{rs_fig_qp_spectra_26_detail}}\hfil
1791: \mbox{\includegraphics[width=0.45\hsize]{rs_fig_qp_spectra_39_detail}}\hfil}
1792: \vspace{1ex}
1793: \hbox to \hsize{\hfil
1794: \hbox to 0.45\hsize{\hfil (c)\hfil}\hfil
1795: \hbox to 0.45\hsize{\hfil (d)\hfil}\hfil}
1796: \vspace{0.5ex}
1797: \hbox to \hsize{\hfil
1798: \mbox{\includegraphics[width=0.45\hsize]{rs_fig_qp_spectra_56_detail}}\hfil
1799: \mbox{\includegraphics[width=0.45\hsize]{rs_fig_qp_spectra_94_detail}}\hfil}
1800: \caption{Fourier spectra of $30\times30$ (blue) and $112\times112$ (red)
1801: approximate 12-fold quasipatterns, as well as the 12-fold quasilattice (small
1802: black dots), up to (a)~26th, (b)~39th, (c)~56th and (d)~94th order.
1803: Here we show wavevectors within $10^\circ$ of $(1,0)$, and with
1804: $0.85\leq|\bfk|\leq1.15$.}
1805: \label{fig:QPFourierSpectraDetail}
1806: \end{figure}
1807:
1808: In figures~\ref{fig:QPFourierSpectra} and~\ref{fig:QPFourierSpectraDetail} we
1809: compare the locations of wavevectors in the $8\times8$ (cyan), $30\times30$
1810: (blue) and $112\times112$ (red) examples from the simulation results shown in
1811: figure~\ref{fig:QPexamples}. The
1812: $8\times8$ spectrum is only in figure~\ref{fig:QPFourierSpectra}(a) (up to 7th
1813: order), and the $30\times30$ spectra are only up to 26th order.
1814: The amplitudes of each Fourier mode is given by the size of the symbol (on a
1815: logarithmic scale): the largest symbols are the modes with the largest
1816: amplitudes, and the modes with the smallest symbols have amplitudes $10^{17}$
1817: times smaller. Modes with amplitudes smaller than this are not plotted as these
1818: are in the realm of round-off error (see below).
1819:
1820: The success with which the periodic patterns approximate a true quasipattern
1821: can be judged by the locations of the important modes in the pattern, as
1822: compared to the locations of modes on the quasilattice, up to a given order. Up
1823: to 7th order (figure~\ref{fig:QPFourierSpectra}a), the $30\times30$ and
1824: $112\times112$ modes (blue and red) overlay each other almost exactly, and
1825: correspond well with the quasilattice modes (small black dots at the centre of
1826: the red markers). However, the $8\times8$ modes (cyan) deviate substantially
1827: from the correct locations, and we conclude that the periodic pattern in an
1828: $8\times8$ domain is a poor approximation to a quasipattern (in spite of
1829: appearances in figure~\ref{fig:QPexamples}a).
1830:
1831: At 11th order (figure~\ref{fig:QPFourierSpectra}b), the agreement between
1832: $30\times30$, $112\times112$ and the true quasipattern is still good, while at
1833: 15th order (figure~\ref{fig:QPFourierSpectra}c), the $30\times30$ modes deviate
1834: noticeably from the $112\times112$ and quasipattern modes. This is more
1835: pronounced at 26th order (figure~\ref{fig:QPFourierSpectraDetail}a).
1836:
1837: At 15th order (figure~\ref{fig:QPFourierSpectra}c), the agreement between
1838: $112\times112$ and the true quasipattern is excellent: every red marker has a
1839: black dot at its centre. The agreement is still very good at 26th order
1840: (figure~\ref{fig:QPFourierSpectraDetail}a): the black dots are not quite in the
1841: centres of the smallest red markers, and some black dots do not have
1842: corresponding red markers. This means that those modes, present in the true
1843: quasipattern at this order, have amplitudes in the $112\times112$ approximation
1844: that are too small to be plotted. The situation at 39th order
1845: (figure~\ref{fig:QPFourierSpectraDetail}b) is similar, but it isn't until 94th
1846: order (figure~\ref{fig:QPFourierSpectraDetail}d) that the true quasipattern
1847: modes miss the centres of the red markers entirely.
1848:
1849: \begin{figure}
1850: \hbox to \hsize{\hfil
1851: \mbox{\includegraphics[width=0.8\hsize]{rs_fig_qp_spectra_qp}}\hfil}
1852: \caption{Fourier spectra of the
1853: $112\times112$ (red) approximate 12-fold quasipattern, as well as the 12-fold
1854: quasilattice (small black dots). All modes with amplitude greater than
1855: $10^{-17}$ times the maximum amplitude are plotted; the quasilattice is plotted
1856: up to 56th order. The competing effects of the 12-fold order and the square
1857: lattice in are apparent.}
1858: \label{fig:QPFourierSpectraLattice}
1859: \end{figure}
1860:
1861: In fact, a curious situation arises at 56th and 94th order: it is apparent that
1862: modes close to $\bfk=(1,0)$, and in particular modes on the line $k_x=1$, have
1863: amplitudes that are higher than might be expected, since usually the amplitudes
1864: of modes decreases with their order. These modes are discussed in more
1865: detail below. However, it should be noted that plotting only modes up to a
1866: certain order masks the effect of the underlying lattice in the numerical
1867: solutions. In figure~\ref{fig:QPFourierSpectraLattice}, we show all modes in
1868: the $112\times112$ down to round-off error, and quasilattice modes up to 56th
1869: order. The underlying square numerical lattice is clearly seen in the strings
1870: of red markers emanating from each large-amplitude mode. These give an
1871: impression that the large-amplitude modes could be considered to be clusters of
1872: modes in Fourier space.
1873:
1874: \begin{figure}
1875: \hbox to \hsize{\hfil (a)\hfil}
1876: \vspace{0.5ex}
1877: \hbox to \hsize{\hfil
1878: \mbox{\includegraphics[width=0.8\hsize]{rs_fig_qp_spectra_order}}\hfil}
1879: \vspace{1ex}
1880: \hbox to \hsize{\hfil (b)\hfil}
1881: \vspace{0.5ex}
1882: \hbox to \hsize{\hfil
1883: \mbox{\includegraphics[width=0.8\hsize]{rs_fig_qp_spectra_order_noise}}\hfil}
1884: \caption{Fourier spectra of $8\times8$ (cyan), $30\times30$ (blue) and
1885: $112\times112$ (red) approximate 12-fold quasipatterns, as a function of
1886: order, (a)~up to 25th order, (b)~up to 100th order ($112\times112$ only). The
1887: vertical lines indicate the range of amplitudes of modes at that order, scaled
1888: to the maximum amplitude at order~1.}
1889: \label{fig:QPFourierSpectraOrder}
1890: \end{figure}
1891:
1892: In figure~\ref{fig:QPFourierSpectraOrder}, we show the range of amplitudes of
1893: the modes as a function of order, for the $8\times8$, $30\times30$ and
1894: $112\times112$ approximate quasipatterns. We note that the $8\times8$ example
1895: deviates significantly from the other two at all orders, while the $30\times30$
1896: and $112\times112$ examples are in fairly good agreement until 19th order or
1897: so, another indication of how the approximation improves with the domain size.
1898: We note that direct comparisons between the $8\times8$ case and the other two
1899: cases are not strictly valid, as the domain size in the
1900: $8\times8$ case had to be adjusted slightly to allow for the different
1901: wavenumbers in the pattern, as discussed above.
1902:
1903: Figure~\ref{fig:QPFourierSpectraOrder}(b) shows the spectrum of the
1904: $112\times112$ example at all orders up to 100. The range of amplitudes at a
1905: given order decays exponentially with order up to about 23rd order, but then
1906: levels off at the level of the round-off error: about $10^{-17}$. On top of
1907: this pattern, there is a peak in amplitude at 15th order, and broad peaks
1908: around 41st, 56th and 97th order. We attribute these to the presence of weakly
1909: damped modes, with $\bfk$ close to unity, that appear around these orders. With
1910: weak damping, these modes will amplify the numerical noise in the PDE
1911: solutions, and nonlinear coupling implies that the amplified noise will feed in
1912: to nearby modes.
1913:
1914: \begin{figure}
1915: \hbox to \hsize{\hfil
1916: \mbox{\includegraphics[width=0.8\hsize]{rs_fig_qp_spectra_closest}}\hfil}
1917: \caption{Minimum value of $\big||\bfkm|-1\big|$, as a function of
1918: order~$|\bfm|$, for $8\times8$ (cyan), $30\times30$ (blue),
1919: $112\times112$ (red) approximate 12-fold quasipatterns, and for a true 12-fold
1920: quasipattern (black)~\cite{Rucklidge2003}. There are drops in this minimum
1921: quantity at 4th, 7th, 11th, 15th, 26th, 39th and 56th order in all cases. Modes
1922: that are responsible for these drops are identified in
1923: figures~\ref{fig:QPFourierSpectra} and~\ref{fig:QPFourierSpectraDetail}.
1924: In the $30\times30$ case, the Pythagorean mode $(0.6,0.8)$ (which is of unit
1925: length) is generated at 26th order.}
1926: \label{fig:QPFourierSpectraClosest}
1927: \end{figure}
1928:
1929: This is shown in figure~\ref{fig:QPFourierSpectraClosest}, where we plot
1930: $\big||\bfkm|-1\big|$ as a function of order~$|\bfm|$ for the different
1931: examples. We note the marked drop in this quantity in particular at 15th and
1932: 39th order. While the drops in $\big||\bfkm|-1\big|$ do not line up exactly
1933: with the peaks in amplitude, we suspect that it is this marked change in the
1934: damping of modes appearing at these orders that is responsible for the
1935: amplification of noise at around the same order.
1936:
1937: Of course, for the numerical patterns in periodic domains, there is a lower
1938: bound to $\big||\bfkm|-1\big|$ that does not depend on~$|\bfm|$ (this lower
1939: bound occurs within figure~\ref{fig:QPFourierSpectraClosest}), while for the
1940: true 12-fold quasipattern, there is no lower bound~\cite{Rucklidge2003}. If
1941: $|\bfkm|\neq1$, then
1942: \begin{equation}
1943: \big||\bfkm|-1\big| > \frac{K}{|\bfm|^2},
1944: \end{equation}
1945: where $K$~is an order-one constant. For some particular
1946: choices of wavevectors, this asymptotic limit is achieved: for example, if
1947: \begin{equation}\label{eq:vectors12}
1948: \bfkm = p \bfkfou + (q-1)\bfknin + (q+1)\bfkele = (1,p-\sqrt{3}q),
1949: \end{equation}
1950: where $p$ and $q$ are integers, then $|\bfm|=p + 2q$ and
1951: $|\bfkm|^2=1+(p-\sqrt{3}q)^2$. When $\frac{p}{q}$ is a
1952: continued fraction
1953: approximation to~$\sqrt{3}$ (table~\ref{table:sqrtthree}), then
1954: $\big||\bfkm|-1\big|\sim\frac{1}{2}(p-\sqrt{3}q)^2\leq\frac{K_2}{q^2}$, so
1955: $\big||\bfkm|-1\big|\leq\frac{K'}{|\bfm|^2}$, where $K_2$ and $K'$ are order-one
1956: constants~\cite{Rucklidge2003}. These particular vectors are not always the
1957: closest ones that can be found at a given order, but they demonstrate that
1958: modes approach the unit circle arbitrarily closely (and so are arbitrarily
1959: weakly damped) as the order of the mode increases. For the fractions listed in
1960: table~\ref{table:sqrtthree}, we have $|\bfm|=p+2q=4$, $11$, $15$, $41$, $56$,
1961: $153$, \dots, but in fact there are step decreases in the minimum of
1962: $\big||\bfkm|-1\big|$ at $|\bfm|=4$, $7$, $11$, $15$, $26$, $39$, $56$, $94$. This is
1963: the reason for the choices of orders in figures~\ref{fig:QPFourierSpectra}
1964: and~\ref{fig:QPFourierSpectraDetail}.
1965:
1966: In summary, having identified which sizes result in the most accurate
1967: approximations to 12-fold quasipatterns, we find that in the largest example
1968: ($112\times112$), the locations of the Fourier modes of the approximation
1969: deviate significantly from those of the true quasipattern only beyond 26th
1970: order. At this point, the amplitudes of the modes have reached the level of
1971: numerical round-off, so going any larger than $112\times112$ would not lead
1972: to any significant improvement in the approximation to a true quasipattern for
1973: these parameter values. The small divisors reveal themselves by amplifying the
1974: amplitudes of the Fourier modes at (or close to) the order at which the small
1975: divisor appears, but they do not appear to cause the amplitudes of the Fourier
1976: modes to become excessively large, at least at this level of forcing, and up to
1977: the maximum order (153) available in this domain.
1978:
1979: % -----------------------------------------------
1980:
1981: \section{Turbulent crystals: quasipatterns using $1:2$ resonance}
1982: \label{sec:Numericsturbulentcrytals}
1983:
1984: In order to have $1:2$ resonance in space and time with single frequency
1985: forcing ($m=1$), we impose ${\hat\Omega}(1)=\frac{1}{2}$ and
1986: ${\hat\Omega}(2)=1$, which leads to $\omega=\frac{1}{3}+4\delta$ and
1987: $\beta=-\frac{1}{6}+5\delta$. We choose $\delta=0$, small values for the
1988: damping coefficients~$\mu$, $\alpha$ and~$\gamma$, and order~one values for the
1989: nonlinear coefficients close to the Hamiltonian limit. The resulting
1990: cross-coupling curve is shown in figure~\ref{fig:BthetaNP}: as explained in
1991: section~\ref{sec:Background}, the $1:2$ resonance in space and time has
1992: enhanced the self-coupling coefficient (by about four orders of magnitude
1993: compared to the previous cases), and so the cross-coupling coefficient
1994: $B_{\theta}$ drops away sharply, and is close to zero for $\theta\geq30^\circ$.
1995:
1996: \begin{figure}
1997: \hbox to \hsize{\hfil
1998: \mbox{\includegraphics[width=0.99\hsize]{rs_fig_np_btheta}}
1999: \hfil}
2000: \caption{Cross-coupling coefficient $B_{\theta}$ for single frequency forcing,
2001: with $1:2$ resonance in space and time: $\omega=\frac{1}{3}$,
2002: $\beta=-\frac{1}{6}$, $\delta=0$, $\mu=-0.005$, $\alpha=0.001$, $\gamma=0$,
2003: $Q_1=3+4i$, $Q_2=-6+8i$, $C=-1+10i$, $F_c=0.024002$ and $k_c=0.9999$.
2004: Note that $B_{\theta}$
2005: drops
2006: away sharply as $\theta$ increases, and is close to zero for
2007: $\theta\geq30^\circ$. The relevant coefficients are
2008: $B_{30}=0.088$, $B_{60}=0.014$ and $B_{90}=0.010$.}
2009: \label{fig:BthetaNP}
2010: \end{figure}
2011:
2012: \begin{figure}
2013: \hbox to \hsize{\hfil
2014: \mbox{\includegraphics[width=0.99\hsize]{rs_fig_np_bifn}}
2015: \hfil}
2016: \caption{Bifurcation diagram showing the weakly nonlinear
2017: predicted amplitude (solid line) and amplitudes of approximate quasipattern
2018: solutions of the PDEs in an $8\times8$ domain (crosses).}
2019: \label{fig:NPbifn}
2020: \end{figure}
2021:
2022: \begin{figure}
2023: \hbox to \hsize{\hfil
2024: \hbox to 0.45\hsize{\hfil (a)\hfil}\hfil
2025: \hbox to 0.45\hsize{\hfil (b)\hfil}\hfil}
2026: \vspace{0.5ex}
2027: \hbox to \hsize{\hfil
2028: \mbox{\includegraphics[width=0.45\hsize]{rs_fig_np30}}\hfil
2029: \mbox{\includegraphics[width=0.45\hsize]{rs_fig_np30_14}}\hfil}
2030: \vspace{1ex}
2031: \hbox to \hsize{\hfil (c)\hfil}
2032: \vspace{0.5ex}
2033: \hbox to \hsize{\hfil
2034: \mbox{\includegraphics[width=0.95\hsize]{rs_fig_np60_20_23}}\hfil}
2035: \caption{For the parameter values from figure~\ref{fig:BthetaNP}, we find three
2036: different approximate quasipatterns depending on the amplitude of the forcing
2037: and the size of the domain.
2038: (a) 1.1 times critical, $30\times30$ domain: a 12-fold quasipattern.
2039: (b) 1.3 times critical, $30\times30$ domain: a 14-fold quasipattern.
2040: (c) 1.3 times critical, $60\times60$ domain: a 20-fold quasipattern
2041: (only $\frac{2}{3}$ of the domain is shown). An animation of the transition
2042: from (a) to (b), also showing details of the Fourier spectrum, can be found
2043: online.}
2044: \label{fig:NPexamples}
2045: \end{figure}
2046:
2047: \begin{figure}
2048: \hbox to \hsize{\hfil
2049: \mbox{\includegraphics[width=0.99\hsize]{rs_fig_np_12_to_14}}
2050: \hfil}
2051: \caption{Amplitudes of Fourier modes as a function of time, at 1.3 times
2052: critical in a $30\times30$ domain. The initial condition was the 12-fold
2053: quasipattern from 1.1 times critical (red). This is unstable and, after an
2054: extended transient of $70\,000$ periods, it is replaced by the 14-fold
2055: quasipattern (green). Amplitudes of other Fourier modes close to $k=1$ are
2056: shown in black. An animation of this transition can be found online.}
2057: \label{fig:NPquasitransition}
2058: \end{figure}
2059:
2060: Within the restrictions of a 12-mode expansion, 12-fold
2061: quasipatterns are stable (figure~\ref{fig:NPbifn}). Indeed, at $1.1$
2062: times the critical forcing, in a $30\times30$ domain, the numerical solution of
2063: the PDE with random initial conditions is a stable 12-fold approximate
2064: quasipattern (figure~\ref{fig:NPexamples}a). As above, the primary modes that
2065: make up the pattern are $(30,0)$ and $(26,15)$ and their reflections, in units
2066: of basic lattice vectors. The amplitudes of the 12~modes differ by $0.5\%$. The
2067: initial condition was not in any invariant subspace, and the PDE was integrated
2068: for $160\,000$ periods of the forcing. The agreement between the weakly
2069: nonlinear predictions and the computed amplitudes is not good
2070: (figure~\ref{fig:NPbifn}), which we expect since the $k=2$ mode is weakly
2071: damped (see discussion above).
2072:
2073: However, there is no feature in the cross-coupling curve
2074: (figure~\ref{fig:BthetaNP}) to indicate that modes at~$30^\circ$ should enjoy a
2075: special status. When the forcing is changed to $1.3$~times critical with this
2076: 12-fold quasipattern as the initial condition, we find that it is unstable, and
2077: is replaced (after a transient of $70\,000$ periods) by a stable approximate
2078: 14-fold quasipattern (figures~\ref{fig:NPexamples}b
2079: and~\ref{fig:NPquasitransition}, and animation online). In this case, the 14
2080: modes are $(30,0)$, $(27,13)$, $(19,23)$, $(7,29)$ and their reflections,
2081: differing in length by $0.5\%$ and having angles within $1.5^\circ$ of
2082: $360^\circ/14$. The amplitudes differ by about $10\%$. Both the 12-fold and
2083: 14-fold quasipatterns are stable at 1.2~times critical, in a $30\times30$
2084: domain.
2085:
2086: \begin{figure}
2087: \hbox to \hsize{\hfil
2088: \mbox{\includegraphics[width=0.99\hsize]{rs_fig_np_20_amplitude}}
2089: \hfil}
2090: \caption{Amplitudes of Fourier modes of the 20-fold quasipattern from
2091: figure~\ref{fig:NPexamples}(c), as a function of wavevector orientation,
2092: for $0.95\leq|\bfk|\leq1.05$, showing twenty peaks roughly evenly distributed.}
2093: \label{fig:NPstructurefunction}
2094: \end{figure}
2095:
2096: More complex quasipatterns are also possible:
2097: calculations done in larger $60\times60$ and $90\times90$ domains at 1.3~times
2098: critical, starting with random initial conditions, both yield an approximate 20-fold
2099: quasipattern (figure~\ref{fig:NPexamples}c). However, this is not a particularly
2100: accurate approximation to a 20-fold quasipattern: in the $60\times60$ case,
2101: the 20~modes are
2102: $(60,0)$,
2103: $(57,18)$,
2104: $(48,36)$,
2105: $(37,47)$,
2106: $(21,56)$,
2107: $(1,60)$,
2108: $(-18,57)$,
2109: $(-33,50)$,
2110: $(-47,37)$,
2111: $(-56,21)$ and their $180^\circ$ rotations, which differ in length by $0.4\%$
2112: and which have angles within $2^\circ$ of $360^\circ/20$. The amplitudes of the
2113: 20 modes differ by up to $40\%$. However, figure~\ref{fig:NPstructurefunction}
2114: shows an examination of the Fourier spectrum of figure~\ref{fig:NPexamples}(c):
2115: there are 20~peaks close to $k=1$, of similar amplitudes and arranged roughly
2116: evenly around the unit circle. The $90\times90$ example is similar.
2117: % the 20 modes are ***still calculating,
2118: % which differ in length by and which have angles within of
2119: % $360^\circ/20$.
2120: We speculate that in other domains and at other forcings, 16-fold and 18-fold
2121: approximate quasipatterns could also be observed.
2122:
2123: The cross-coupling $B_\theta$ curve suggests that modes that are more than
2124: about $30^\circ$ apart do not influence each other (at least in an amplitude
2125: equation truncated at cubic order). This suggests that patterns containing many
2126: modes at essentially arbitrary angles might be expected -- such patterns have
2127: been termed {\em turbulent crystals} by Newell and Pomeau~\cite{Newell1993}. It
2128: is hard to see why one quasipattern should be favoured over another, on indeed,
2129: why quasipatterns (with wavevectors evenly distributed around the $k=1$ circle)
2130: should be favoured over more complex patterns.
2131:
2132: We have not attempted to compare the locations of the high-order modes in these
2133: 14- and 20-fold approximate quasipatterns with those in the true quasipatterns,
2134: as these higher quasilattices are very densely populated
2135: (figure~\ref{fig:lattices}d), and the approximate solutions are not close
2136: enough to the true quasipatterns.
2137:
2138: As an aside, there is an interesting connection that can be made between the
2139: Fourier spectra of these high-order quasipatterns and the fractal dynamics of
2140: the complex ODE $d^2\zeta/dt^2=-\zeta^{n-1}$, where $n$~is an even
2141: integer~\cite{Grinevich2007}. Solutions of this equation lie on multiply
2142: branched Riemann surfaces, with branch points occurring densely at points in a
2143: quasilattice of order~$n$.
2144:
2145: The rotational degeneracy of the plane (in the absence of
2146: boundary conditions) implies that any
2147: mode with $|\bfk|=1$ is linearly excited, and it is an open question as to why
2148: patterns and quasipatterns, with a finite number of modes evenly distributed
2149: around the unit circle, should be the preferred patterns close to onset in so
2150: many examples of pattern-forming systems. (Of course, pattern-forming systems
2151: are by definition those that produce regular patterns close to onset!) Newell
2152: and Pomeau~\cite{Newell1993} wrote down the evolution equation of $N$ modes,
2153: ignoring quadratic terms and truncating at cubic order:
2154: \begin{equation}
2155: \frac{dA_j}{dt} = \lambda A_j - \sum_{k=1}^{N} B_{\theta_{jk}} |A_k|^2A_j,
2156: \end{equation}
2157: where $A_j$ is the complex amplitude of mode~$j$, $\lambda$ is the growth rate,
2158: and $\theta_{jk}$ is the angle between the wavevectors of modes $j$ and~$k$.
2159: Truncated in this way, the phases of the amplitudes do not enter the dynamics:
2160: Newell and Pomeau~\cite{Newell1993} attribute this to the translation symmetry
2161: of the underlying problem, but Melbourne~\cite{Melbourne1998} points out that
2162: this phase-invariance is in fact a normal form symmetry and hence not exact.
2163:
2164: This system of ODEs is variational, and evolves to minimise a free energy
2165: \begin{equation}
2166: \label{eq:ZVLF}
2167: \mathcal{F} = -\lambda\sum_{j=1}^{N}|A_j|^2
2168: + \frac{1}{2}\sum_{j,k=1}^{N} B_{\theta_{jk}} |A_j|^2|A_k|^2.
2169: \end{equation}
2170: Newell and Pomeau~\cite{Newell1993} claim that in the case that the state that
2171: minimizes the free energy~$\mathcal{F}$ has many modes, with the magnitudes of
2172: all amplitudes equal but with arbitrary phases, then the pattern will resemble
2173: a spatially random field (a turbulent crystal). This case will be realised when
2174: $B_{\theta}<1$ over a wide range of~$\theta$ (as in figure~\ref{fig:BthetaNP},
2175: for example). We therefore propose the quasipattern solutions in
2176: figure~\ref{fig:NPexamples} (12-, 14- and 20-fold quasipatterns) as examples of
2177: turbulent crystals.
2178:
2179: \begin{figure}
2180: \hbox to \hsize{\hfil
2181: \mbox{\includegraphics[width=0.8\hsize]{rs_fig_zvlf}}\hfil}
2182: \caption{Value of the free energy~$\mathcal{F}$ (\ref{eq:ZVLF}) for $N$-fold
2183: patterns and quasipatterns, derived from the $B_\theta$ curves in
2184: figure~\ref{fig:BthetaNP} with single-frequency forcing (blue) and
2185: figure~\ref{fig:BthetaQP} with multi-frequency forcing (red).}
2186: \label{fig:ZVLF}
2187: \end{figure}
2188:
2189: It is interesting to consider what determines the number of modes in these
2190: turbulent crystals. One argument, explored in more detail by Zhang and
2191: Vi\~nals~\cite{Zhang1997}, is that the preferred pattern at onset should be the
2192: global minimum of the free energy~$\mathcal{F}$. Assuming equal amplitudes, the
2193: free energy of an $N$-fold pattern or quasipattern depends on $B_{\theta}$
2194: evaluated at~$360^\circ/N$, $720^\circ/N$ {\em etc.} Figure~\ref{fig:ZVLF}
2195: shows $\mathcal{F}$ derived from the $B_\theta$ curves in
2196: figures~\ref{fig:BthetaNP} and \ref{fig:BthetaQP}, with single and
2197: multi-frequency forcing respectively.
2198:
2199: In the single-frequency case (blue curve in figure~\ref{fig:ZVLF}), where
2200: $B_{\theta}$ is close to zero for $\theta\geq30^\circ$
2201: (figure~\ref{fig:BthetaNP}), there is a broad minimum around 14-fold or 16-fold
2202: quasipatterns. However, 12-fold, 14-fold and 20-fold examples were found
2203: (figure~\ref{fig:NPexamples}), depending on the forcing strength and domain
2204: size. No doubt other patterns could also be found with more exploration.
2205:
2206: In the multi-frequency case (red curve in figure~\ref{fig:ZVLF}), where
2207: $B_{\theta}$ has pronounced dips at $30^\circ$ and $90^\circ$
2208: (figure~\ref{fig:BthetaQP}), there are local minima at $N=4$ (squares), $N=8$
2209: and $N=12$. In numerical experiments, most initial conditions found squares,
2210: though 12-fold quasipatterns were also stable (figures~\ref{fig:QPbifn}
2211: and~\ref{fig:QPexamples}). We have not looked for 8-fold quasipatterns.
2212:
2213: These results suggest that these free energy arguments provide a useful
2214: qualitative tool for understanding pattern selection, but reality is often more
2215: complicated than the arguments might suggest.
2216:
2217: % -----------------------------------------------
2218:
2219: \section{Conclusions}
2220: \label{sec:Conclusions}
2221:
2222: We have introduced a new model PDE~(\ref{eq:pde}) for investigating pattern
2223: formation and pattern design in parametrically forced systems. The PDE is
2224: intended to play the same role for the Faraday wave experiment that the
2225: Swift--Hohenberg equation~\cite{Swift1977} plays for convection: while the
2226: model cannot be derived from the fluid mechanics, it has qualitatively correct
2227: linear behaviour and the right type of nonlinear interactions in order to
2228: provide useful illumination of the processes that are going on. The model
2229: produces superlattice patterns (section~\ref{sec:Numericssuperlattice}) and
2230: quasipatterns (sections~\ref{sec:Numericsquasipatterns}
2231: and~\ref{sec:Numericsturbulentcrytals}) in response to single and
2232: multi-frequency forcing, for the same reasons that these complex patterns are
2233: found in the Faraday wave experiment -- confirming that the mechanisms are
2234: generic. The ease of calculating weakly nonlinear coefficients and of computing
2235: large-scale numerical solutions has allowed a quantitative exploration of the
2236: agreement between the theoretical understanding of the pattern selection
2237: mechanism and the patterns that are actually found.
2238:
2239: Of course, the model does not capture every detail of the physics of the
2240: Faraday wave experiment. In particular, the dispersion relations have different
2241: structures, and the model does not include the mean flow effects that are
2242: important in nearly inviscid Faraday experiments~\cite{Knobloch2002}. The
2243: latter could be addressed by coupling the model to a conserved quantity or to a
2244: mean-flow equation (as in~\cite{Tsimring1997,Greenside1985}) or by taking the
2245: negative Laplacian of the right-hand side of the PDE (as
2246: in~\cite{Cox2003,Dawes2003}).
2247:
2248: The Zhang--Vi\~nals~\cite{Zhang1996} equations do not have these drawbacks:
2249: these are derived from the Navier--Stokes equations in the limit of infinite
2250: depth and zero viscosity. One might ask what is gained by looking at a simpler
2251: PDE that is even further from the physics. There are two advantages of the new
2252: model: one is that it is very simple: the dispersion relation can be controlled
2253: easily for studying any resonant interaction or response to multi-frequency
2254: forcing; in addition, the weakly nonlinear theory can be computed very easily.
2255: A second advantage is that is is very well suited to the use of efficient
2256: numerical methods such as Exponential Time Differencing~\cite{Cox2002}: the
2257: linear terms are diagonal in Fourier space, and the nonlinear terms do not
2258: involve any derivatives. In contrast, the Zhang--Vi\~nals equations are
2259: considerably more complicated and the weakly nonlinear computations are more
2260: involved. Numerical solutions are also more time-consuming, as the linear term
2261: is not diagonal in spectral space, and most of the nonlinear terms involve
2262: products of derivatives, resulting in more Fourier transforms for their
2263: evaluation. As a result of the relatively low cost of the calculations, we have
2264: been able to follow branches of solutions in detail, and to go to much larger
2265: domains and for much longer times than previous calculations. Of course, in the
2266: end it would be desirable to work directly with the Navier--Stokes equations,
2267: but for these, the weakly nonlinear theory is very
2268: challenging~\cite{Skeldon2007} and there are as yet no large-scale numerical
2269: simulations.
2270:
2271: Like the Zhang--Vi\~nals equations, the new model includes explicit time
2272: dependence. In contrast, other approaches, based on developing a description of
2273: the slow evolution of the amplitude of an underlying pattern, use the
2274: Ginzburg--Landau equation with additional complex conjugate terms to capture
2275: the effect of the time-dependent
2276: forcing~\cite{Coullet1992c,Conway2007,Conway2007a,Halloy2007}. As a result, any
2277: complex patterns that are found must be interpreted in terms of slow,
2278: long-wavelength amplitude modulations of an underlying pattern, which
2279: complicates any effort to make quantitative comparison between theoretical
2280: ideas and the behaviour of the real system.
2281:
2282: The significance of three-wave coupling to weakly damped modes and its role in
2283: pattern selection has long been
2284: recognised~\cite{Mermin1985,Newell1993,Edwards1994,Zhang1997}. We have put this
2285: idea to a quantitative test by using it to choose forcing functions that
2286: stabilise a desired pattern in large domain calculations. However, the
2287: codimension-one approach to this idea, where the weakly damped modes are slaved
2288: to the pattern-forming modes, does not provide quantitative predictions of
2289: amplitudes of patterns, and of parameter regimes where the desired patterns
2290: should be stable, except for very close to onset. The reason is that
2291: computation of the cross-coupling coefficient~$B_{\theta}$ is only valid when
2292: all modes are strongly damped compared to the pattern-forming modes. This poses
2293: difficulties because the most interesting patterns occur where the
2294: pattern-forming modes are coupled to weakly damped modes, and this is where the
2295: theory used to calculate properties of these patterns is of limited validity.
2296: As a result, parameters had to be chosen very close to onset in order to find
2297: stable numerical examples of the desired patterns in parameter regimes where
2298: they were predicted to be stable. A codimension-two approach would extend the
2299: range of validity of the theory, and will be the subject of future work.
2300:
2301: We have investigated two mechanisms for the formation of quasipatterns. One
2302: mechanism uses three-wave interactions involving a damped mode associated with
2303: the difference of the two frequencies in the forcing to select a particular
2304: angle ($30^\circ$~in the example presented here). Using different primary
2305: frequencies, or altering the dispersion relation, would allow other angles, or
2306: combinations of angles, to be selected. The advantage is that a forcing
2307: function can be designed to produce a particular pattern: the mechanism is
2308: quite selective, and requires some fine-tuning of the parameters.
2309:
2310: The second mechanism uses $1:2$ resonance in space and time to magnify the
2311: self-interaction coefficient and thereby, on rescaling, diminish the
2312: cross-coupling coefficient~$B_{\theta}$ for angles greater than
2313: about~$30^\circ$. This can lead to the formation of turbulent
2314: crystals~\cite{Newell1993}. The mechanism is robust (the patterns are found
2315: well above onset), and requires only single frequency forcing. A dispersion
2316: relation that supports $1:2$ resonance in space and time is needed. Within this
2317: framework, an inherent complication is that it is not clear why regular 8, 10,
2318: 12 or 14-fold quasipatterns, or indeed any other combination of modes, should
2319: be favoured. Indeed we have found that 12-, 14- and 20-fold approximate
2320: quasipatterns can be stabilised by altering the level of the forcing or the
2321: domain size, without changing other parameters, and we have reported the
2322: transition between two different types of quasipattern. The Lyapunov function
2323: approach~\cite{Zhang1996} cannot make this distinction, and would predict that
2324: 14- or 16-fold quasipatterns should be found at onset for these parameter
2325: values (figure~\ref{fig:ZVLF}). It remains an open question as to why one
2326: turbulent crystal should be favoured over another.
2327:
2328: The existence of 14-fold (and higher) quasipatterns has been suggested
2329: before~\cite{Zhang1996,Rucklidge2003,Topaz2004,Steurer2007}, but we have
2330: presented here the first examples of spontaneously formed 14-fold and 20-fold
2331: approximate quasipatterns that are stable solution of a PDE
2332: (figure~\ref{fig:NPexamples}), with preliminary results
2333: in~\cite{Rucklidge2007}. Examples where 14-fold symmetry is imposed externally
2334: have been reported in optical experiments~\cite{Pampaloni1995}. The Fourier
2335: spectra of 12-fold and 14-fold quasipatterns are both dense
2336: (figure~\ref{fig:lattices}c,d), but those of 14-fold quasipatterns are much
2337: denser, owing to the difference between quadratic and cubic irrational
2338: numbers~\cite{Rucklidge2003}. This difference may have profound consequences
2339: for their mathematical treatment.
2340:
2341: We have identified what domain sizes result in the most accurate approximations
2342: to 12-fold quasipatterns, based on square and on hexagonal domains
2343: (table~\ref{table:domains}), and produced exceptionally clean examples of
2344: approximate quasipatterns in relatively large computational domains. Comparing
2345: the Fourier spectra of the approximate quasipatterns as a function of domain
2346: size, we have identified at what order the locations of Fourier modes in the
2347: approximate quasipatterns deviate from those of the true quasipatterns. In the
2348: largest example ($112\times112$), the locations of the Fourier modes deviate
2349: significantly only beyond 26th order
2350: (figure~\ref{fig:QPFourierSpectraDetail}a), at which point the amplitudes of
2351: the modes have reached the level of numerical round-off
2352: (figure~\ref{fig:QPFourierSpectraOrder}). This suggests that going any larger
2353: than $112\times112$ would not lead to any significant improvement in the
2354: approximation to a true quasipattern, at least for these parameter values.
2355:
2356: We have compared the amplitudes of the Fourier modes of the approximate
2357: quasipatterns and the leading order weakly nonlinear prediction
2358: (figures~\ref{fig:QPbifn} and~\ref{fig:NPbifn}), and found quantitative
2359: agreement very close to onset, but only qualitative agreement at larger
2360: amplitude, which is what would be expected from the problem of eliminating
2361: weakly damped modes, as discussed above. We have not extended this comparison
2362: to higher order since the weakly nonlinear calculations are too difficult for
2363: this parametrically forced problem. An extension of this work would be to
2364: devise a PDE without parametric forcing that also produces stable quasipattern
2365: solutions: this would allow high order weakly nonlinear calculations (as
2366: in~\cite{Rucklidge2003}) and very large domain numerical solutions, and so allow
2367: a comparison between computed mode amplitudes (as a function of order) and the
2368: weakly nonlinear theory. Standard weakly nonlinear theory produces amplitudes
2369: that diverge at high order because of the presence of small
2370: divisors~\cite{Rucklidge2003}, while the PDE solutions have amplitudes that
2371: decay exponentially with order -- although the small divisors in this problem
2372: do make themselves felt by amplifying the magnitudes of the Fourier modes at (or
2373: close to) the order at which the small divisor appears
2374: (figure~\ref{fig:QPFourierSpectraOrder}). Such a PDE could be based on (for
2375: example) the Swift--Hohenberg
2376: equation~\cite{Frisch1995,Lifshitz1997,Muller1994}, but the Swift--Hohenberg
2377: equation itself does not allow the weakly damped modes that are necessary to
2378: stabilise quasipatterns.
2379:
2380: Other numerical studies of quasipatterns as solutions of a PDE have not made a
2381: systematic study of the effect of domain size. Zhang and
2382: Vi\~nals~\cite{Zhang1996,Zhang1998} report approximate 8-fold quasipatterns in
2383: a $64\times64$ domain in their quasipotential model of the nearly inviscid
2384: Navier--Stokes equations, for parameter values close to the $1:2$ resonance in
2385: space and time. The modes involved were separated by $41^\circ$, $42^\circ$,
2386: $48^\circ$ and $49^\circ$~\cite{Zhang1998}, so the approximation was not
2387: particularly accurate; our careful choice of domain size allowed much closer
2388: approximation. M\"uller~\cite{Muller1994} developed a model based on two
2389: coupled Swift--Hohenberg equations, with parameters chosen so that the two
2390: unstable modes had wavenumbers that would favour 8-fold or 12-fold
2391: quasipatterns. Numerical simulations in a $10\times10$ domain in the second
2392: case found approximate 12-fold quasipatterns. The modes involved are not
2393: stated, but we estimate them to be $\bfkone=(10,-1)$, $\bfktwo=(9,4)$,
2394: $\bfkthr=(6,8)$ in units of the fundamental lattice vector. These have lengths
2395: $10.05$, $9.85$ and $10.00$ respectively, and they are separated by
2396: $28.48^\circ$ and $29.17^\circ$. With $\bfkele=(4,-9)$, we have
2397: $\bfkthr+\bfkele=\bfkone$, so the $60^\circ$~resonance condition is satisfied.
2398: Frisch and Sonnino\cite{Frisch1995} present a similar model and report 10-fold
2399: quasipatterns. Lifshitz and Petrich~\cite{Lifshitz1997} found a 12-fold
2400: approximate quasipattern in a roughly $30\times30$ domain, in a model based on
2401: a single Swift--Hohenberg equation with a degenerate double minimum in its
2402: marginal stability curve. The modes involved appear to be the same as those in
2403: the $30\times30$ examples discussed in section~\ref{sec:Numericsquasipatterns}.
2404:
2405: While we have not discussed the possibility of long-wave instabilities of
2406: quasipatterns, the Fourier spectra of the $112\times112$
2407: example~(figures~\ref{fig:QPFourierSpectraDetail}(d)
2408: and~\ref{fig:QPFourierSpectraLattice}) suggests that long-wave modes that are
2409: close to the primary wavevectors in a direction {\em tangent} to the critical
2410: circle are forced by high-order nonlinear interactions. This is also apparent
2411: from the locations of modes responsible for the small
2412: divisors~\cite{Rucklidge2003}. Therefore, any treatment of the long-wave
2413: stability of quasipatterns should take into account the presence of these
2414: modes. This is a delicate question. The only study of the sideband
2415: instabilities of quasipatterns~\cite{Echebarria2001} focusses on instabilities
2416: associated with modes that are perpendicular to the unit circle, using coupled
2417: Ginzburg--Landau equations for each of the primary mode directions in the
2418: quasipattern. This approach could be extended to include instabilities
2419: associated with modes that are tangent to the unit circle by looking at coupled
2420: Newell--Whitehead--Segel equations, along the lines suggested
2421: by~\cite{Gunaratne1994}, although high-order nonlinear interaction may not be
2422: captured in a long-wave analysis truncated at cubic order.
2423:
2424: % ------------------------------------------
2425:
2426: \section*{Acknowledgments}
2427: We are grateful for support from National Science Foundation (DMS-0309667) and
2428: from the Engineering and Physical Sciences Research Council (GR/S45928/01). We
2429: are also grateful to many people who have helped shape these ideas: Jessica
2430: Conway, St\'ephan Fauve, Jay Fineberg, Rebecca Hoyle, G\'erard Iooss, Edgar
2431: Knobloch, Paul Matthews, Ian Melbourne, Werner Pesch, Michael Proctor, Jeff
2432: Porter, Hermann Riecke, Anne Skeldon, Jorge Vi\~nals and Gene Wayne. We thank
2433: Michael Proctor for pointing out the effect of higher order terms on the centre
2434: manifold, discussed in section~\ref{sec:Numericssuperlattice}. Finally, we are
2435: grateful to the Isaac Newton Institute for Mathematical Sciences, where part of
2436: this work was carried out.
2437:
2438: % ------------------------------------------
2439:
2440: \begingroup
2441:
2442: \def\url#1{}
2443:
2444: % this is for bibtex
2445: \bibliographystyle{siam}
2446: \bibliography{rs}
2447:
2448: \endgroup
2449:
2450: % ------------------------------------------
2451:
2452: \appendix
2453:
2454: \section{Weakly nonlinear theory}
2455: \label{app:WNLT}
2456:
2457: In this appendix, we present the weakly nonlinear theory for the
2458: PDE~(\ref{eq:pde}). We will describe the calculation in terms of a harmonic
2459: primary bifurcation; the subharmonic case is similar, with the main differences
2460: being that the period is $4\pi$ rather then~$2\pi$, and that the quadratic
2461: coefficient $Q$ is identically zero.
2462:
2463: We start by writing $U=u+iv$, where $u(x,y,t)$ and $v(x,y,t)$ are real
2464: functions, and so
2465: \begin{align*}
2466: \frac{\partial u}{\partial t}&=
2467: \left(\mu+\alpha\nabla^2+\gamma\nabla^4\right)u
2468: -\left(\omega+\beta\nabla^2+\delta\nabla^4\right)v\\
2469: &\qquad{}
2470: +Q_{1r}(u^2-v^2) - Q_{1i}(2uv) + Q_{2r}(u^2+v^2)
2471: +C_r(u^2+v^2)u - C_i(u^2+v^2)v\\
2472: \frac{\partial v}{\partial t}&=
2473: \left(\omega+\beta\nabla^2+\delta\nabla^4\right)u
2474: +\left(\mu+\alpha\nabla^2+\gamma\nabla^4\right)v\\
2475: &\qquad{}
2476: +Q_{1i}(u^2-v^2) + Q_{1r}(2uv) + Q_{2i}(u^2+v^2)
2477: +C_r(u^2+v^2)v + C_i(u^2+v^2)u\\
2478: &\qquad{}+f(t)u.
2479: \end{align*}
2480: We define differential operators $\mathcal L$ and $\mathcal M$:
2481: \begin{equation*}
2482: {\mathcal L}=\frac{\partial}{\partial t}-\left(\mu+\alpha\nabla^2+\gamma\nabla^4\right)
2483: \qquad\hbox{and}\qquad
2484: {\mathcal M}=\left(\omega+\beta\nabla^2+\delta\nabla^4\right),
2485: \end{equation*}
2486: so the PDEs for $u$ and $v$ are
2487: \begin{align*}
2488: {\mathcal L} u &= - {\mathcal M}v + \hbox{NL}_u,\\
2489: {\mathcal L} v &= \phantom{-}{\mathcal M}u + \hbox{NL}_v + f(t)u.
2490: \end{align*}
2491: The nonlinear terms $\hbox{NL}_u$ and $\hbox{NL}_v$ are:
2492: \begin{align*}
2493: \hbox{NL}_u &=
2494: Q_{1r}(u^2-v^2) - Q_{1i}(2uv) + Q_{2r}(u^2+v^2)
2495: +C_r(u^2+v^2)u - C_i(u^2+v^2)v,\\
2496: \hbox{NL}_v &=
2497: Q_{1i}(u^2-v^2) + Q_{1r}(2uv) + Q_{2i}(u^2+v^2)
2498: +C_r(u^2+v^2)v + C_i(u^2+v^2)u.
2499: \end{align*}
2500:
2501: \subsection{Linear theory}
2502:
2503: We seek solutions of the form $u=\eikdotx p_1(t)$ and $v=\eikdotx q_1(t)$, where $p_1$ and $q_1$ are
2504: periodic functions of period~$T$, and define
2505: \begin{equation*}
2506: {\hat\gamma}_1=2\left(-\mu+\alpha k^2-\gamma k^4\right),
2507: \qquad
2508: {\hat\Omega}_1=\omega-\beta k^2+\delta k^4,
2509: \qquad
2510: \Omega_1=\sqrt{\left(\frac{{\hat\gamma}_1}{2}\right)^2 + \left({\hat\Omega}_1\right)^2},
2511: \end{equation*}
2512: we get
2513: \begin{align*}
2514: {\mathcal L}_1 p_1 &= - {\mathcal M}_1 q_1,\\
2515: {\mathcal L}_1 q_1 &= \phantom{-}{\mathcal M}_1 p_1 + f(t)p_1,
2516: \end{align*}
2517: or
2518: \begin{equation*}
2519: {\mathcal L}_1^2 p_1 = - {\mathcal M}_1{\mathcal L}_1 q_1
2520: = - {\mathcal M}_1^2 p_1 - f(t){\mathcal M}_1 p_1,
2521: \end{equation*}
2522: where ${\mathcal L}_1$ and ${\mathcal M}_1$ act on $p_1(t)$ and $q_1(t)$ as
2523: \begin{equation*}
2524: {\mathcal L}_1=\frac{d}{dt}+\frac{{\hat\gamma}_1}{2}
2525: \qquad\hbox{and}\qquad
2526: {\mathcal M}_1={\hat\Omega}_1.
2527: \end{equation*}
2528: The linearised PDE reduces to a damped Mathieu equation for $p_1$:
2529: \begin{equation*}
2530: \left(\frac{d}{dt}+\frac{{\hat\gamma}_1}{2}\right)^2 p_1 + {\hat\Omega}_1^2 p_1 + f(t){\hat\Omega}_1 p_1=0
2531: \end{equation*}
2532: or
2533: \begin{equation*}
2534: {\ddot p}_1 + {\hat\gamma}_1 {\dot p}_1 + \left(\Omega_1^2 + {\hat\Omega}_1 f(t)\right) p_1=0
2535: ={\mathbf L}p_1.
2536: \end{equation*}
2537: The adjoint equation is:
2538: \begin{equation*}
2539: {\ddot {\tilde p}}_1 - {\hat\gamma}_1{\dot {\tilde p}}_1 + \left(\Omega_1^2 + {\hat\Omega}_1f(t)\right){\tilde p}_1=0
2540: ={\mathbf{\tilde L}}{\tilde p}_1,
2541: \end{equation*}
2542: with respect to an inner product
2543: \begin{equation*}
2544: \langle g,h\rangle=\frac{1}{T}\int_0^T\, g(t)h(t)\,dt,
2545: \end{equation*}
2546: with $T=2\pi$ (harmonic case) or $T=4\pi$ (subharmonic case),
2547: so $\langle g,{\mathbf L}h\rangle= \langle {\mathbf{\tilde L}}g,h\rangle$
2548:
2549: For a given value of~$k$, seeking periodic solutions of ${\mathbf L}p=0$ yields
2550: an eigenvalue problem whose eigenvalue is the amplitude of the forcing function
2551: $f(t)$. We use the method of~\cite{Besson1996} to solve this eigenvalue problem
2552: with multi-frequency forcing~$f(t)$, providing the critical forcing amplitude.
2553: Minimising this critical forcing amplitude over $k$ yields the critical
2554: wavenumber $k_c$, critical forcing function $f_c(t)$, and critical
2555: eigenfunction $p_1(t)$. The corresponding $q_1(t)$ is determined by solving
2556: ${\mathcal L}_1q_1={\hat\Omega}p_1+f_c(t)p_1$.
2557:
2558: \subsection{Rhombs}
2559:
2560: We consider $f$ close to $f_c$,
2561: writing $f(t)=f_c(t) (1+\epsilon^2F_2)$,
2562: and seek small-amplitude rhombic solutions associated with two wavevectors
2563: $\bfkone$ and $\bfktwo$ at the critical wavenumber:
2564: $k_1=k_2=k_c$, separated by an angle~$\theta$. We formally expand the solution as
2565: \begin{align*}
2566: u&=\epsilon u_1 + \epsilon^2 u_2 + \epsilon^3 u_3 + \cdots\\
2567: {}&=\epsilon\left(
2568: z_1(T_2)\eikonedotx + z_2(T_2)\eiktwodotx + \cc
2569: \right)p_1(t)\\
2570: &\quad{}
2571: + \epsilon^2\Big(
2572: \left(z_1^2\etwoikonedotx + z_2^2\etwoiktwodotx + \cc\right)p_2(t)
2573: + \left(|z_1|^2 + |z_2|^2\right)p_3(t)\\
2574: &\quad\qquad{}
2575: + \left(z_1z_2\eikonepktwodotx + \cc\right)p_4(t)
2576: + \left(z_1{\bar z}_2\eikonemktwodotx + \cc\right)p_5(t)
2577: \Big)
2578: + {\mathcal O}(\epsilon^3),
2579: \end{align*}
2580: with a similar expression for $v$ in terms of $v_1$, $v_2$ and $v_3$, and
2581: $q_1$, ..., $q_5$, where $T_2$ is a slow time, varying on a scale of
2582: $\epsilon^{-2}$, and the functions $p_2(t)$, ..., $q_5(t)$ are to be
2583: determined. The form of this solution is chosen by knowing in advance the
2584: structure of the nonlinear terms and the modes to be generated by them.
2585:
2586: Substituting these expressions for $u$ and $v$ into the PDE and ordering
2587: in powers of~$\epsilon$, we recover,
2588: at leading order in~$\epsilon$, the linear theory.
2589: At second order in~$\epsilon$, we split the PDE into terms that go as
2590: $\etwoikonedotx$, $\etwoiktwodotx$, terms without spatial dependence,
2591: and terms that go as $\eikonepktwodotx$ and $\eikonemktwodotx$.
2592:
2593: The terms like $\etwoikonedotx$ lead to equations for~$p_2$ and~$q_2$:
2594: \begin{align*}
2595: {\mathcal L}_2 p_2 &= - {\mathcal M}_2 q_2 + \hbox{NL}^{(2)}_{p2},\\
2596: {\mathcal L}_2 q_2 &= \phantom{-}{\mathcal M}_2 p_2 + \hbox{NL}^{(2)}_{q2} + f_c(t)p_2,
2597: \end{align*}
2598: where the linear operators are:
2599: \begin{align*}
2600: {\mathcal L}_2&=\frac{d}{dt}-\left(\mu-4\alpha k_c^2+16\gamma k_c^4\right)
2601: =\frac{d}{dt}+\frac{{\hat\gamma}_2}{2},\\
2602: {\mathcal M}_2&=\left(\omega-4\beta k_c^2+16\delta k_c^4\right)={\hat\Omega}_2,
2603: \end{align*}
2604: and the nonlinear terms
2605: $\hbox{NL}^{(2)}_{p2}$ and $\hbox{NL}^{(2)}_{q2}$ are:
2606: \begin{align*}
2607: \hbox{NL}^{(2)}_{p2} &=
2608: Q_{1r}(p_1^2-q_1^2) - Q_{1i}(2p_1q_1) + Q_{2r}(p_1^2+q_1^2),\\
2609: \hbox{NL}^{(2)}_{q2} &=
2610: Q_{1i}(p_1^2-q_1^2) + Q_{1r}(2p_1q_1) + Q_{2i}(p_1^2+q_1^2).
2611: \end{align*}
2612: The function $q_2$ is eliminated by operating with ${\mathcal L}_2$, resulting
2613: in a second-order non-constant coefficient inhomogeneous linear ODE for~$p_2$:
2614: \begin{equation*}
2615: \left({\mathcal L}_2^2 + {\mathcal M}_2^2 + {\mathcal M}_2f(t)\right)p_2 =
2616: {\mathcal L}_2 \hbox{NL}^{(2)}_{p2} - {\mathcal M}_2 \hbox{NL}^{(2)}_{q2}.
2617: \end{equation*}
2618: or
2619: \begin{equation*}
2620: {\ddot p_2} + {\hat\gamma}_2 {\dot p_2} + \left(\Omega_2^2 + {\hat\Omega}_2f_c(t)\right)p_2 =
2621: \left(\frac{d}{dt} + \frac{{\hat\gamma}_2}{2}\right)\hbox{NL}^{(2)}_{p2} - {\hat\Omega}_2 \hbox{NL}^{(2)}_{q2}.
2622: \end{equation*}
2623: This can be solved numerically for $p_2$ using Fourier transform methods, and
2624: $q_2$ can then be found. Terms that go as $\etwoiktwodotx$ result in the same
2625: equation.
2626:
2627: Terms without spatial dependence, and terms with spatial dependence $\eikonepktwodotx$ and $\eikonemktwodotx$,
2628: result in similar equations for $p_3$, $p_4$ and $p_5$, but with linear operators:
2629: \begin{align*}
2630: {\mathcal L}_3&=\frac{d}{dt}- \mu = \frac{d}{dt} + \frac{{\hat\gamma}_3}{2},
2631: \qquad\hbox{and}\qquad
2632: {\mathcal M}_3=\omega={\hat\Omega}_3,\\
2633: {\mathcal L}_4&=\frac{d}{dt}-\left(\mu-4\cos^2\left(\frac{\theta}{2}\right)\alpha k_c^2+16\cos^4\left(\frac{\theta}{2}\right)\gamma k_c^4\right)
2634: = \frac{d}{dt} + \frac{{\hat\gamma}_4}{2},\\
2635: {\mathcal M}_4&=\left(\omega-4\cos^2\left(\frac{\theta}{2}\right)\beta k_c^2+16\cos^4\left(\frac{\theta}{2}\right)\delta k_c^4\right)
2636: = {\hat\Omega}_4,\\
2637: {\mathcal L}_5&=\frac{d}{dt}-\left(\mu-4\sin^2\left(\frac{\theta}{2}\right)\alpha k_c^2+16\sin^4\left(\frac{\theta}{2}\right)\gamma k_c^4\right)
2638: = \frac{d}{dt} + \frac{{\hat\gamma}_5}{2},\\
2639: {\mathcal M}_5&=\left(\omega-4\sin^2\left(\frac{\theta}{2}\right)\beta k_c^2+16\sin^4\left(\frac{\theta}{2}\right)\delta k_c^4\right)
2640: = {\hat\Omega}_5,
2641: \end{align*}
2642: and nonlinear terms:
2643: \begin{align*}
2644: \hbox{NL}^{(2)}_{p3} &= \hbox{NL}^{(2)}_{p4} = \hbox{NL}^{(2)}_{p5} = 2\hbox{NL}^{(2)}_{p2},\\
2645: \hbox{NL}^{(2)}_{q3} &= \hbox{NL}^{(2)}_{q4} = \hbox{NL}^{(2)}_{q5} = 2\hbox{NL}^{(2)}_{q2}.
2646: \end{align*}
2647: Note that ${\hat\gamma}_4$, ${\hat\gamma}_5$, ${\hat\Omega}_4$ and ${\hat\Omega}_5$ depend on~$\theta$, the angle
2648: between the chosen wavevectors, whereas ${\hat\gamma}_1$, ${\hat\gamma}_2$, ${\hat\gamma}_3$, ${\hat\Omega}_1$,
2649: ${\hat\Omega}_2$ and ${\hat\Omega}_3$ do not.
2650:
2651: At third order in~$\epsilon$, the problem has the following structure:
2652: \begin{align*}
2653: {\mathcal L} u_3 + \frac{\partial u_1}{\partial T_2} &= - {\mathcal M}v_3 + \hbox{NL}^{(3)}_u,\\
2654: {\mathcal L} v_3 + \frac{\partial v_1}{\partial T_2} &= \phantom{-}{\mathcal M}u_3 + \hbox{NL}^{(3)}_v + f_c(t)u_3 + F_2f_c(t)u_1.
2655: \end{align*}
2656: The nonlinear terms $\hbox{NL}^{(3)}_u$ and $\hbox{NL}^{(3)}_v$ are:
2657: \begin{align*}
2658: \hbox{NL}^{(3)}_u &=
2659: 2Q_{1r}(u_1 u_2 - v_1 v_2) - Q_{1i}(2u_1v_2+2u_2v_1) + 2Q_{2r}(u_1u_2+v_1v_2)\\
2660: &\qquad{}+C_r(u_1^2+v_1^2)u_1 - C_i(u_1^2+v_1^2)v_1,\\
2661: \hbox{NL}^{(3)}_v &=
2662: 2Q_{1i}(u_1 u_2 - v_1 v_2) + Q_{1r}(2u_1v_2+2u_2v_1) + 2Q_{2i}(u_1u_2+v_1v_2)\\
2663: &\qquad{}+C_r(u_1^2+v_1^2)v_1 + C_i(u_1^2+v_1^2)u_1.
2664: \end{align*}
2665: Eliminating~$v_3$, we obtain
2666: \begin{equation*}
2667: \left({\mathcal L}^2+{\mathcal M}^2+{\mathcal M}f_c\right) u_3 =
2668: -{\mathcal L}\frac{\partial u_1}{\partial T_2}
2669: +{\mathcal M}\frac{\partial v_1}{\partial T_2}
2670: -{\mathcal M}F_2f_cu_1
2671: +{\mathcal L}\hbox{NL}^{(3)}_u - {\mathcal M}\hbox{NL}^{(3)}_v.
2672: \end{equation*}
2673: The operator on the left is the singular operator from the linearised problem,
2674: so the equation can only be solved for $u_3$ if a solvability condition is
2675: applied to the terms that are proportional to $\eikonedotx$ and $\eiktwodotx$
2676: and complex conjugates. If we take the inner product between ${\tilde p}_1$ and
2677: the $\eikonedotx$ component of the above, we find \begin{equation*}
2678: \tau\frac{\partial z_1}{\partial T_2} =
2679: \sigma F_2 z_1
2680: + \left({\hat A} |z_1|^2 + ({\hat B}_{\hbox{indep}}+{\hat B}_{\hbox{res}}(\theta))|z_2|^2\right)z_1,
2681: \end{equation*}
2682: with a similar equation for $z_2$, where
2683: \begin{equation*}
2684: \tau=\left\langle {\tilde p}_1,
2685: 2 \left({\dot p}_1 + \frac{{\hat\gamma}_1}{2}p_1\right)
2686: \right\rangle
2687: \qquad\hbox{and}\qquad
2688: \sigma=\left\langle {\tilde p}_1,
2689: -{\hat\Omega}_1 f_c p_1
2690: \right\rangle
2691: \end{equation*}
2692: (using ${\mathcal L}_1p_1=-{\mathcal M}_1q_1$) and
2693: \begin{align*}
2694: {\hat A} &= \Big\langle {\tilde p}_1,
2695: \Big(\frac{d}{dt}+\frac{{\hat\gamma}_1}{2}\Big)
2696: \Big(2Q_{1r}(p_1 p_2 + p_1 p_3 - q_1 q_2 - q_1 q_3) \\
2697: & \qquad\qquad\qquad\qquad
2698: {} - 2Q_{1i}(p_1 q_2 + p_1 q_3 + q_1 p_2 + q_1 p_3) \\
2699: & \qquad\qquad\qquad\qquad
2700: {} + 2Q_{2r}(p_1 p_2 + p_1 p_3 + q_1 q_2 + q_1 q_3) \\
2701: & \qquad\qquad\qquad\qquad
2702: {} + 3C_r(p_1^2+q_1^2)p_1
2703: - 3C_i(p_1^2+q_1^2)q_1\Big)\\
2704: & \qquad {} - {\hat\Omega}_1
2705: \Big(2Q_{1r}(p_1 q_2 + p_1 q_3 + q_1 p_2 + q_1 p_3) \\
2706: & \qquad\qquad\qquad
2707: {} + 2Q_{1i}(p_1 p_2 + p_1 p_3 - q_1 q_2 - q_1 q_3) \\
2708: & \qquad\qquad\qquad
2709: {} + 2Q_{2i}(p_1 p_2 + p_1 p_3 + q_1 q_2 + q_1 q_3) \\
2710: & \qquad\qquad\qquad
2711: {} + 3C_r(p_1^2+q_1^2)q_1
2712: + 3C_i(p_1^2+q_1^2)p_1\Big)
2713: \Big\rangle, \\
2714: {\hat B}_{\hbox{indep}} &= \Big\langle {\tilde p}_1,
2715: \Big(\frac{d}{dt}+\frac{{\hat\gamma}_1}{2}\Big)
2716: \Big(2Q_{1r}(p_1 p_3 - q_1 q_3)
2717: - 2Q_{1i}(p_1 q_3 + q_1 p_3) \\
2718: & \qquad\qquad\qquad\qquad
2719: {} + 2Q_{2r}(p_1 p_3 + q_1 q_3)
2720: + 6C_r(p_1^2+q_1^2)p_1
2721: - 6C_i(p_1^2+q_1^2)q_1\Big)\\
2722: & \qquad {} - {\hat\Omega}_1
2723: \Big(2Q_{1r}(p_1 q_3 + q_1 p_3)
2724: + 2Q_{1i}(p_1 p_3 - q_1 q_3)
2725: + 2Q_{2i}(p_1 p_3 + q_1 q_3) \\
2726: & \qquad\qquad\qquad
2727: {} + 6C_r(p_1^2+q_1^2)q_1
2728: + 6C_i(p_1^2+q_1^2)p_1\Big)
2729: \Big\rangle,\\
2730: {\hat B}_{\hbox{res}}(\theta) &= \Big\langle {\tilde p}_1,
2731: \Big(\frac{d}{dt}+\frac{{\hat\gamma}_1}{2}\Big)
2732: \Big(2Q_{1r}(p_1 p_4 + p_1 p_5 - q_1 q_4 - q_1 q_5) \\
2733: & \qquad\qquad\qquad\qquad
2734: {} - 2Q_{1i}(p_1 q_4 + p_1 q_5 + q_1 p_4 + q_1 p_5) \\
2735: & \qquad\qquad\qquad\qquad
2736: {} + 2Q_{2r}(p_1 p_4 + p_1 p_5 + q_1 q_4 + q_1 q_5)\Big)\\
2737: & \qquad {} - {\hat\Omega}_1
2738: \Big(2Q_{1r}(p_1 q_4 + p_1 q_5 + q_1 p_4 + q_1 p_5) \\
2739: & \qquad\qquad\qquad
2740: {} + 2Q_{1i}(p_1 p_4 + p_1 p_5 - q_1 q_4 - q_1 q_5) \\
2741: & \qquad\qquad\qquad
2742: {} + 2Q_{2i}(p_1 p_4 + p_1 p_5 + q_1 q_4 + q_1 q_5)\Big)
2743: \Big\rangle.
2744: \end{align*}
2745: For convenience, we have separated the parts of the cross-coupling coefficient
2746: that do not depend on the angle between the modes (${\hat B}_{\hbox{indep}}$)
2747: from those that do (${\hat B}_{\hbox{res}}(\theta)$). We discuss below how
2748: these coefficients are then scaled.
2749:
2750: \subsection{Hexagons}
2751:
2752: As with rhombs, we look for $f$ close to $f_c$, writing $f(t)=f_c(t)
2753: (1+\epsilon^2F_2)$, but now we chose three wavevectors, $\bfkone$, $\bfktwo$
2754: and $\bfkthr$ oriented at $120^\circ$ to each other, with
2755: $\bfkone+\bfktwo+\bfkthr=\mathbf{0}$. We look for small-amplitude solutions
2756: with equal amplitudes of the three waves, and write
2757: \begin{align*}
2758: u&=\epsilon u_1 + \epsilon^2 u_2 + \epsilon^3 u_3 + \cdots\\
2759: {}&=\epsilon z(T_1,T_2)\left(
2760: \eikonedotx + \eiktwodotx + \eikthrdotx + \cc
2761: \right)p_1(t)\\
2762: &\quad{}
2763: + \epsilon^2\Big(
2764: z^2\left(\etwoikonedotx + \etwoiktwodotx + \etwoikthrdotx + \cc\right)p_2(t)
2765: + 3|z|^2 p_3(t)\\
2766: &\quad\qquad{}
2767: + |z|^2 \left(\eikonemktwodotx +\eiktwomkthrdotx +\eikthrmkonedotx + \cc\right){\tilde p}_5(t)\\
2768: &\quad\qquad{}
2769: + {\bar z}^2 \left(
2770: \eikonedotx + \eiktwodotx + \eikthrdotx + \cc
2771: \right)p_6(t)
2772: \Big)
2773: + {\mathcal O}(\epsilon^3),
2774: \end{align*}
2775: with a similar expression for $v$ in terms of $q_1$, ..., $q_6$, where $T_1$
2776: and $T_2$ are slow times, varying on scales $\epsilon^{-1}$ and
2777: $\epsilon^{-2}$, and the functions $p_2(t)$, ..., $q_6(t)$ are to be
2778: determined. The form of the expression is chosen by knowing in advance the
2779: structure of the nonlinear terms.
2780:
2781: Substituting these expressions for $u$ and $v$ into the PDE and ordering in
2782: powers of~$\epsilon$, at leading order in~$\epsilon$ we recover the linear
2783: theory. At second order in~$\epsilon$, we split the PDE into terms that go as
2784: $\etwoikonedotx+\etwoiktwodotx+\etwoikthrdotx+\cc$, terms without spatial
2785: dependence, terms that go as $\eikonemktwodotx +\eiktwomkthrdotx
2786: +\eikthrmkonedotx + \cc$, and finally terms that go as $\eikonedotx +
2787: \eiktwodotx + \eikthrdotx + \cc$, which have to be considered specially.
2788:
2789: Terms like $\etwoikonedotx+\etwoiktwodotx+\etwoikthrdotx+\cc$ and terms without
2790: spatial dependence give the same equations for $p_2$, $q_2$, $p_3$ and $q_3$ as
2791: in the case of rhombs. In particular, the inhomogeneous nonlinear terms are the
2792: same. Terms that go as $\eikonemktwodotx +\eiktwomkthrdotx +\eikthrmkonedotx +
2793: \cc$ result in equations for ${\tilde p}_5$ and ${\tilde q}_5$ that are the
2794: equations for $p_5$ and $q_5$ evaluated for $\theta=120^\circ$.
2795:
2796: Terms that go as $\eikonedotx + \eiktwodotx + \eikthrdotx + \cc$ require the
2797: use of two time-scales and a solvability condition. The linear operators are
2798: the same as for the initial linear problem:
2799: \begin{align*}
2800: {\mathcal L}_1 p_6 &= - {\mathcal M}_1 q_6 + \hbox{NL}^{(2)}_{p6} \phantom{{}+ f_c(t)p_6}
2801: - \frac{\partial z/\partial T_1}{{\bar z}^2} p_1,\\
2802: {\mathcal L}_1 q_6 &= \phantom{-}{\mathcal M}_1 p_6 + \hbox{NL}^{(2)}_{q6} + f_c(t)p_6
2803: - \frac{\partial z/\partial T_1}{{\bar z}^2} q_1.
2804: \end{align*}
2805: The nonlinear terms are
2806: $\hbox{NL}^{(2)}_{p6}$ and $\hbox{NL}^{(2)}_{q6}$ are:
2807: \begin{equation*}
2808: \hbox{NL}^{(2)}_{p6} = 2\hbox{NL}^{(2)}_{p2}
2809: \qquad\hbox{and}\qquad
2810: \hbox{NL}^{(2)}_{q6} = 2\hbox{NL}^{(2)}_{q2}.
2811: \end{equation*}
2812: This can be reduced to a
2813: second-order non-constant coefficient inhomogeneous linear ODE for~$p_6$:
2814: \begin{equation*}
2815: \left({\mathcal L}_1^2 + {\mathcal M}_1^2 + {\mathcal M}_1f(t)\right)p_6 =
2816: {\mathcal L}_1 \hbox{NL}^{(2)}_{p6} - {\mathcal M}_1 \hbox{NL}^{(2)}_{q6}
2817: + \frac{\partial z/\partial T_1}{{\bar z}^2}
2818: \left({\mathcal M}_1 q_1 - {\mathcal L}_1 p_1\right)
2819: \end{equation*}
2820: Since the operator on the LHS is the singular linear operator~${\mathbf L}$, we
2821: must apply a solvability condition:
2822: \begin{equation*}
2823: \langle {\tilde p}_1,{\mathbf L}p_6\rangle = 0 =
2824: \langle {\tilde p}_1,
2825: {\mathcal L}_1 \hbox{NL}^{(2)}_{p6} - {\mathcal M}_1
2826: \hbox{NL}^{(2)}_{q6} \rangle
2827: - 2 \frac{\partial z/\partial T_1}{{\bar z}^2}
2828: \langle {\tilde p}_1,
2829: {\mathcal L}_1 p_1
2830: \rangle
2831: \end{equation*}
2832: since ${\mathcal M}_1 q_1=-{\mathcal L}_1 p_1$. We define
2833: \begin{equation*}
2834: \tau=\left\langle{\tilde p}_1,
2835: 2\left(\frac{dp_1}{dt}+\frac{{\hat\gamma}_1}{2}p_1\right)
2836: \right\rangle
2837: \end{equation*}
2838: as before and
2839: \begin{equation*}
2840: {\hat\epsilon}=\left\langle{\tilde p}_1,
2841: {\mathcal L}_1 \hbox{NL}^{(2)}_{p6} -
2842: {\mathcal M}_1 \hbox{NL}^{(2)}_{q6}
2843: \right\rangle,
2844: \end{equation*}
2845: and obtain an equation for the slow ($T_1$) evolution of the amplitude~$z$:
2846: \begin{equation*}
2847: \tau \frac{\partial z}{\partial T_1} = {\hat\epsilon} {\bar z}^2.
2848: \end{equation*}
2849: Once the solvability condition has been imposed, the ODE can be solved for $p_6$ and
2850: $q_6$. The computed solution~$p_6$ contains an arbitrary amount of~$p_1$; the solution
2851: is made unique by specifying that $\langle{\tilde p}_1,p_6\rangle=0$.
2852:
2853: At third order in~$\epsilon$, the problem has the following structure:
2854: \begin{align*}
2855: {\mathcal L} u_3 + \frac{\partial u_2}{\partial T_1} + \frac{\partial u_1}{\partial T_2} &=
2856: - {\mathcal M}v_3 + \hbox{NL}^{(3h)}_u,\\
2857: {\mathcal L} v_3 + \frac{\partial v_2}{\partial T_1} + \frac{\partial v_1}{\partial T_2} &=
2858: \phantom{-}{\mathcal M}u_3 + \hbox{NL}^{(3h)}_v + f_c(t)u_3 + F_2f_c(t)u_1.
2859: \end{align*}
2860: We only need keep track of terms proportional to $\eikonedotx$ in our
2861: derivation of the bifurcation problem, so the $\partial u_2/\partial T_1$ and
2862: $\partial v_2/\partial T_1$ terms yield $p_6\partial {\bar z}^2/\partial T_1$
2863: and $q_6\partial {\bar z}^2/\partial T_1$. The $\eikonedotx$ components of the
2864: nonlinear terms $\hbox{NL}^{(3h)}_u$ and $\hbox{NL}^{(3h)}_v$ are specified
2865: below.
2866:
2867: Eliminating~$v_3$, we obtain
2868: \begin{align*}
2869: \left({\mathcal L}^2+{\mathcal M}^2+{\mathcal M}f_c\right) u_3 &=
2870: -{\mathcal L}\left(\frac{\partial u_2}{\partial T_1} + \frac{\partial u_1}{\partial T_2}
2871: - \hbox{NL}^{(3h)}_u \right)\\
2872: &\qquad{}
2873: +{\mathcal M}\left(\frac{\partial v_2}{\partial T_1} + \frac{\partial v_1}{\partial T_2}
2874: - F_2f_cu_1 - \hbox{NL}^{(3h)}_v \right).
2875: \end{align*}
2876: The equation can only be solved for $u_3$ if a solvability condition
2877: is applied to the terms that are proportional to $\eikonedotx$,
2878: $\eiktwodotx$ and $\eikthrdotx$, and complex conjugates. If we take the inner product
2879: between ${\tilde p}_1$ and the
2880: $\eikonedotx$ component of the above, we find
2881: \begin{equation*}
2882: \tau\frac{\partial z}{\partial T_2} =
2883: \sigma F_2 z
2884: + \left({\hat A}+2{\hat B}_{60}\right) |z|^2 z,
2885: \end{equation*}
2886: where $\tau$ and $\sigma$ are unchanged from the rhombic calculations, and
2887: \begin{align*}
2888: {\hat A}+2{\hat B}_{60} &= \Big\langle {\tilde p}_1,
2889: \Big(\frac{d}{dt}+\frac{{\hat\gamma}_1}{2}\Big)\hbox{NL}^{(3h)}_u -
2890: {\hat\Omega}_1 \hbox{NL}^{(3h)}_v\\
2891: & \qquad\qquad {}+\left(-{\mathcal L}_1 p_6 + {\mathcal M}_1 q_6\right)2\frac{{\hat\epsilon}}{\tau}
2892: \Big\rangle,
2893: \end{align*}
2894: where we have used
2895: \begin{equation*}
2896: \frac{\partial{\bar z}^2}{\partial T_1} = 2{\bar z}\frac{\partial{\bar z}}{\partial T_1}
2897: = 2\frac{{\hat\epsilon}}{\tau} |z|^2 z
2898: \end{equation*}
2899: and
2900: \begin{align*}
2901: \hbox{NL}^{(3h)}_u &=
2902: 2Q_{1r}(p_1p_2 + 3p_1p_3 + 2p_1{\tilde p}_5 + 2p_1p_6
2903: - q_1q_2 - 3q_1q_3 - 2q_1{\tilde q}_5 - 2q_1q_6)\\
2904: & \qquad
2905: {} - 2Q_{1i}(p_1q_2 + 3p_1q_3 + 2p_1{\tilde q}_5 + 2p_1q_6
2906: + q_1p_2 + 3q_1p_3 + 2q_1{\tilde p}_5 + 2q_1p_6)\\
2907: & \qquad
2908: {} + 2Q_{1r}(p_1p_2 + 3p_1p_3 + 2p_1{\tilde p}_5 + 2p_1p_6
2909: + q_1q_2 + 3q_1q_3 + 2q_1{\tilde q}_5 + 2q_1q_6)\\
2910: & \qquad
2911: {} + 15C_r(p_1^2+q_1^2)p_1 - 15C_i(p_1^2+q_1^2)q_1,\\
2912: \hbox{NL}^{(3h)}_v &=
2913: 2Q_{1r}(p_1q_2 + 3p_1q_3 + 2p_1{\tilde q}_5 + 2p_1q_6
2914: + q_1p_2 + 3q_1p_3 + 2q_1{\tilde p}_5 + 2q_1p_6)\\
2915: & \qquad
2916: {} + 2Q_{1i}(p_1p_2 + 3p_1p_3 + 2p_1{\tilde p}_5 + 2p_1p_6
2917: - q_1q_2 - 3q_1q_3 - 2q_1{\tilde q}_5 - 2q_1q_6)\\
2918: & \qquad
2919: {} + 2Q_{2i}(p_1p_2 + 3p_1p_3 + 2p_1{\tilde p}_5 + 2p_1p_6
2920: + q_1q_2 + 3q_1q_3 + 2q_1{\tilde q}_5 + 2q_1q_6)\\
2921: & \qquad
2922: {} + 15C_r(p_1^2+q_1^2)q_1 + C_i(p_1^2+q_1^2)p_1.
2923: \end{align*}
2924: From the value of $\hat A$ calculated for rhombs, we can recover ${\hat
2925: B}_{60}$, which is effectively the cross-coupling coefficient for modes with
2926: wavevectors at~$60^\circ$.
2927:
2928: \subsection{Reconstitution}
2929:
2930: At this stage, the pattern formation problem on a hexagonal lattice would take
2931: the form:
2932: \begin{align*}
2933: \tau\frac{\partial z_1}{\partial T_1} &= {\hat \epsilon} {\bar z}_2 {\bar z}_3,\\
2934: \tau\frac{\partial z_1}{\partial T_2} &=
2935: \sigma F_2 z_1
2936: + \left({\hat A} |z_1|^2 + {\hat B}_{60}|z_2|^2+{\hat B}_{60}|z_3|^2\right)z_1,
2937: \end{align*}
2938: where $z_1$, $z_2$ and $z_3$ are amplitudes of $\eikonedotx$, $\eiktwodotx$ and
2939: $\eikthrdotx$. Similar equations are found for $\partial z_2/\partial T_1$ {\em
2940: etc.} Recall that the small factor~$\epsilon$ has been used so that the
2941: amplitude of the original amplitude~$U$ is explicitly small: $U=\epsilon
2942: z_1(T_1,T_2)\eikonedotx p_1(t)+\dots$
2943:
2944: There is more than one way to combine these equations into a single ODE.
2945: Properly, we should consider only the case where the coefficient of the
2946: quadratic term $\hat\epsilon$ is itself small (order~$\epsilon$). This occurs
2947: either for small values of $Q_1$ and~$Q_2$, or
2948: near a codimension-one line in $(Q_1,Q_2)$ space:
2949: \begin{align*}
2950: 0 &= {\hat\epsilon}=\left\langle{\tilde p}_1,
2951: {\mathcal L}_1 \hbox{NL}^{(2)}_{p6} -
2952: {\mathcal M}_1 \hbox{NL}^{(2)}_{q6}\right\rangle\\
2953: &= 2Q_{1r}\left\langle{\tilde p}_1,
2954: {\mathcal L}_1 (p_1^2-q_1^2) -
2955: {\mathcal M}_1 (2p_1q_1)\right\rangle +
2956: 2Q_{1i}\left\langle{\tilde p}_1,
2957: {\mathcal L}_1 (-2p_1q_1) -
2958: {\mathcal M}_1 (p_1^2-q_1^2)\right\rangle \\
2959: &\qquad\qquad {}+ 2Q_{2r}\left\langle{\tilde p}_1,
2960: {\mathcal L}_1 (p_1^2+q_1^2)\right\rangle +
2961: 2Q_{2i}\left\langle{\tilde p}_1,
2962: - {\mathcal M}_1 (p_1^2+q_1^2)\right\rangle.
2963: \end{align*}
2964: Alternatively, we note that for many of these multi-frequency forced problems,
2965: ${\hat\epsilon}$ is small anyway~\cite{Porter2002,Porter2004a}.
2966:
2967: Having decided that ${\hat\epsilon}$ is small,
2968: we define a fast time scale~$t^*$, related to the original time scale~$t$
2969: by an order-one factor $\sigma/\tau$:
2970: \begin{equation*}
2971: \frac{d}{dt^*}=\epsilon\frac{\tau}{\sigma} \frac{\partial}{\partial T_1} +
2972: \epsilon^2\frac{\tau}{\sigma} \frac{\partial}{\partial T_2},
2973: \end{equation*}
2974: where $\epsilon$ is the original small parameter. We now scale the $z$'s by
2975: $1/\epsilon$, so that the original amplitude~$U$ is implicitly small:
2976: $U=z_1(T_1,T_2)\eikonedotx p_1(t)+\dots$, and obtain:
2977: \begin{equation*}
2978: \frac{dz_1}{dt^*} = \lambda z_1 + \frac{{\hat \epsilon}}{\sigma} {\bar z}_2 {\bar z}_3
2979: + \frac{{\hat A}}{\sigma}\left(|z_1|^2 + B_{60}|z_2|^2+B_{60}|z_3|^2\right)z_1,
2980: \end{equation*}
2981: where $\lambda=\epsilon^2 F_2$, so that the forcing amplitude is $(1+\lambda)$
2982: times the critical amplitude, and $B_{60}={\hat B}_{60}/{\hat A}$.
2983:
2984: The advantage of reconstituting in this way is that the quadratic and cubic
2985: terms appear at the same order. The disadvantage is that the regime of validity
2986: ($\lambda\ll1$, ${\hat\epsilon}\ll1$, $z\ll1$) is not made explicit. In
2987: particular, this validity condition will only be satisfied for hexagons when
2988: the coefficient of the quadratic term is small -- which is precisely the limit
2989: required for the quadratic and cubic terms to be of the same order.
2990:
2991: Finally, we scale the amplitudes once more by a factor of $\sqrt{|\sigma/{\hat
2992: A}|}$, rename the time variable back to~$t$, and obtain:
2993: \begin{equation*}
2994: \frac{dz_1}{dt} = \lambda z_1 + Q {\bar z}_2 {\bar z}_3
2995: + s \left(|z_1|^2 + B_{60}|z_2|^2+B_{60}|z_3|^2\right)z_1,
2996: \end{equation*}
2997: with similar equations for $z_2$ and $z_3$,
2998: where
2999: \begin{equation*}
3000: Q =\frac{{\hat \epsilon}}{\sigma}
3001: \sqrt{\left|\frac{\sigma}{{\hat A}}\right|}
3002: \qquad\hbox{and}\qquad
3003: s=\hbox{sgn}\left(\frac{{\hat A}}{\sigma}\right)
3004: \end{equation*}
3005: (usually, $s=-1$).
3006:
3007: Repeating the same reconstitution for the rhombic lattice results in
3008: \begin{align*}
3009: \frac{dz_1}{dt} &= \lambda z_1
3010: + s \left(|z_1|^2 + B_{\theta}|z_2|^2\right)z_1\\
3011: \frac{dz_2}{dt} &= \lambda z_2
3012: + s \left(|z_2|^2 + B_{\theta}|z_1|^2\right)z_2
3013: \end{align*}
3014: where $B_{\theta}=({\hat B}_{\hbox{indep}} + {\hat B}_{\hbox{res}}(\theta))/{\hat A}$.
3015:
3016: The equations for other lattices can be found from combinations of the above,
3017: as can candidate equations for quasipatterns.
3018:
3019: \end{document}
3020:
3021:
3022: