1: \documentclass[11pt]{amsart}
2: \usepackage{amsmath}
3: \usepackage{amssymb}
4: \usepackage{epsfig}
5: \usepackage{a4wide}
6: %% \usepackage[normalem]{ulem}
7:
8:
9: \usepackage{pifont}
10: \newcommand{\todo}[1]{[\textnormal{\ding{46}\ding{55}} #1]}
11:
12: \newcommand{\B}{\ensuremath{\mathbb{B}}}
13: \newcommand{\N}{\ensuremath{\mathbb{N}}}
14: \newcommand{\Z}{\ensuremath{\mathbb{Z}}}
15: \newcommand{\R}{\ensuremath{\mathbb{R}}}
16: \newcommand{\T}{\ensuremath{{\mathcal T}}}
17: \newcommand{\F}{\ensuremath{{\mathcal F}}}
18: \newcommand{\bs}[1]{\boldsymbol{#1}}
19: \newcommand{\E}{\ensuremath{\mathbb{E}}}
20: \newcommand{\X}{\ensuremath{\boldsymbol X}}
21:
22:
23: \DeclareMathOperator{\inn}{int}
24: \DeclareMathOperator{\cl}{cl}
25: \DeclareMathOperator{\supp}{supp}
26: \DeclareMathOperator{\dens}{dens}
27: \DeclareMathOperator{\conv}{conv}
28:
29:
30: \newtheorem{thm}{Theorem}[section]
31: \newtheorem{lemma}[thm]{Lemma}
32: \newtheorem{prop}[thm]{Proposition}
33: \newtheorem{defi}[thm]{Definition}
34: % \newtheorem{corollary}[thm]{Corollary}
35:
36: \parskip1ex
37: \parindent0em
38:
39: \begin{document}
40:
41: \title{Duality of Model Sets generated by Substitutions}
42:
43: \author{D.~Frettl\"oh}
44:
45: \address{Fakult\"at f\"ur Mathematik, Universit\"at
46: Bielefeld, Postfach 100131, 33501 Bielefeld, Germany}
47: \email{dirk.frettloeh@math.uni-bielefeld.de}
48: \urladdr{http://www.math.uni-bielefeld.de/baake/frettloe}
49:
50: \maketitle
51:
52: \begin{center}
53: {\em Dedicated to Tudor Zamfirescu \\
54: on the occasion of his sixtieth birthday}
55: \end{center}
56:
57:
58: \begin{abstract}
59: The nature of this paper is twofold: On one hand, we will give a short
60: introduction and overview of the theory of model sets in connection
61: with nonperiodic substitution tilings and generalized Rauzy fractals.
62: On the other hand, we will construct certain Rauzy fractals and a
63: certain substitution tiling with interesting properties, and we will
64: use a new approach to prove rigorously that the latter one arises
65: from a model set.
66: The proof will use a duality principle which will be described in
67: detail for this example. This duality is mentioned as early as
68: 1997 in \cite{gel} in the context of iterated function systems, but it
69: seems to appear nowhere else in connection with model sets.
70: \end{abstract}
71:
72:
73: \section{Introduction\label{intro}}
74: One of the essential observations in the theory of nonperiodic, but
75: highly ordered structures is the fact that, in many cases, they can be
76: generated by a certain projection from a higher dimensional (periodic)
77: point lattice. This is true for the well-known Penrose tilings (cf.~
78: \cite{gs}) as well as for a lot of other substitution tilings, where
79: most known examples are living in $\E^1,\E^2$ or $\E^3$. The
80: appropriate framework for this arises from the theory of model sets
81: (\cite{mey}, \cite{moo}). Though the knowledge in this field has grown
82: considerably in the last decade, in many cases it is still hard to
83: prove rigorously that a given nonperiodic structure indeed is a model
84: set.
85:
86: This paper is organized as follows. Section \ref{sec:subst} briefly
87: collects some well--known facts about substitution
88: tilings. These facts will be used later. Section
89: \ref{sec:modset} contains some basics about model sets, in particular
90: in connection with substitution tilings. In Section 1.3 we give the
91: definition of Rauzy fractals. This can be done completely in the
92: framework of model sets; it is just a question of the terminology.
93: % In Section \ref{sec:zwo} we describe the duality
94: % principle by using a certain example.
95: In Section \ref{sec:ansatz} we introduce a family of one--dimensional
96: substitutions which are generalizations of the well--known Fibonacci
97: sequences. Four of these possess a substitution factor which is a
98: PV--number, so these are candidates for being model sets. Section
99: \ref{sec:n4} is dedicated to one of these substitutions. Therein, the
100: corresponding Rauzy fractal is obtained. Section \ref{sec:dual} shows
101: the construction of the dual substitution tiling, which is a
102: two--dimensional tiling with some interesting properties. Using the
103: duality, we will show that this dual tiling is a model set. Hence,
104: this is one of the few known cases of a two--dimensional model set,
105: generated by a substitution, where the substitution factor is
106: not an algebraic number of degree one or degree two (and the proof
107: being rigorous). Section \ref{sec:remarks} contains some additional
108: remarks, and further examples of dual substitutions, where the duality
109: principle can be used in order to prove that these are model sets, too.
110:
111:
112: Let us fix some notation. Throughout the paper, $\cl(M)$ denotes the
113: closure of a set $M$, $\inn(M)$ the interior of $M$, $\partial M$ the
114: boundary of $M$, and $\#M$ the cardinality of $M$. $\E^d$ denotes the
115: $d$--dimensional Euclidean space, i.e., $\R^d$ equipped with the
116: Euclidean metric $\| \cdot \|$. $\B^d$ denotes the unit ball $\{ {\bs
117: x} \, | \, \|{\bs x}\|\le 1 \}$. If not stated otherwise, $\mu$
118: denotes the $d$--dimensional Lebesgue measure, where $d$ will be clear
119: from the context.
120: A \emph{tile} is a nonempty compact set $T \subset \E^d$ with the property
121: $\cl(\inn(T))=T$. A \emph{tiling} of $\E^d$ is a collection of tiles
122: $\{T^{}_n\}^{}_{n \ge 0}$ which covers $\E^d$ and contains no
123: {\em overlapping} tiles, i.e., $\inn(T^{}_k) \cap \inn(T^{}_n) =
124: \varnothing$ for $k \ne n$. In other words: A tiling is both a
125: covering and a packing of $\E^d$. A tiling $\T$ is called
126: \emph{nonperiodic}, if the only solution of $\T+\bs{x}=\T$ is
127: $\bs{x}=\bs{0}$.
128:
129: \subsection{Substitution Tilings} \label{sec:subst}
130: A convenient method to generate nonperiodic tilings is by a
131: {\em substitution}: Choose a set of {\em prototiles}, that is, a set of
132: tiles $\F:=\{T_1, T_2, \ldots, T_m\}$. Choose a
133: {\em substitution factor} $\lambda \in \R, \lambda > 1$ and a rule how to
134: dissect $\lambda T_i$ (for every $1 \le i \le m$) into tiles, such that
135: any of these tiles is congruent to some tile $T_j \in \F$. It will be
136: sufficient for this paper to think of a substitution like in
137: Fig.~\ref{inflbsp}. For completeness, we give the proper definition
138: here.
139:
140:
141: \begin{defi} \label{substdef}
142: Let $\F:=\{T_1, T_2, \ldots, T_m\}$ be a set of tiles, the
143: {\em prototiles}, and $\lambda>1$ a real number, the
144: {\em substitution factor}. Let $\lambda T_i = \varphi_1(T_{i_1}) \cup
145: \varphi_2(T_{i_2}) \cup \ldots \cup \varphi_{n(i)}(T_{i_{n(i)}})$,
146: such that every $\varphi_j$ is an isometry of $\E^d$ and the
147: involved tiles do not overlap. Furthermore, let
148: $\sigma(\{T_i\})= \{ \varphi_1(T_{i_1}), \varphi_2(T_{i_2}), \ldots,
149: \varphi_{n(i)}(T_{i_{n(i)}}) \}$. Let ${\bs S}$ be the set of all
150: configurations of tiles congruent to tiles in $\F$. That is, ${\bs S}$ is
151: the set of all sets of the form $\{T_{j_i} +{\bs x}_i \, | \, T_{j_i} \in
152: \F, i \in I, {\bs x}_i \in \E^d \}$, where $I$ is some
153: finite or infinite index set. By the requirement
154: $\sigma(\{T_i+{\bs x}\})=\sigma(\{T_i\})+\lambda {\bs x}$ (${\bs x}
155: \in \E^d$), $\sigma$ extends in a
156: unique way to a well--defined map from ${\bs S}$ to ${\bs S}$.
157: Then $\sigma: {\bs S} \to {\bs S}$ is called a {\em substitution}.
158: Every tiling of $\E^d$, where any finite part of it is congruent to a
159: subset of some $\sigma^k(T_i)$ is called a
160: {\em substitution tiling (with substitution $\sigma$)}.
161: \end{defi}
162:
163: % In particular,
164: % $\sigma$ can be iterated on a single tile. The resulting configuration
165: % of tiles cover larger and larger areas of $\E^d$.
166: More general, this definition makes sense if we replace $\lambda$
167: by some expanding linear map. That is, replace $\lambda$ in the
168: definition above by a matrix $Q$ such that all eigenvalues of $Q$ are
169: larger than 1 in modulus. In this full generality, many of the following
170: results may break down. Anyway, in Section \ref{sec:n4} we will use the
171: generalization in one certain case, where it will cause no problem.
172:
173: To a substitution $\sigma$ we assign a {\em substitution matrix}:
174: $A_{\sigma}:=(a_{ij})_{1 \le i,j \le m}$, where $a_{ij}$ is the number
175: of tiles $T_j+{\bs x}$ (tiles of type $T_j \in \F$) in $\sigma(\{T_i\})$.
176: In this paper we will consider substitutions with {\em primitive}
177: substitution matrices only. Recall: A nonnegative matrix $A$ is called
178: {\em primitive}, if some power $A^k$ is a strictly positive matrix. In
179: this situation we can apply the following well--known theorem.
180:
181: \begin{thm}[Perron--Frobenius] \label{perron}
182: Let $A \in \R^{m \times m}$ be a primitive nonnegative matrix.
183: Then $A$ has an eigenvalue $0< \lambda \in \R$, which is simple and
184: larger in modulus than every other eigenvalue of $A$. This
185: eigenvalue is called {\em Perron--Frobenius--eigenvalue} or shortly
186: {\em PF--eigenvalue}. \\
187: The corresponding eigenvector can be chosen such that it is strictly
188: positive. Such an eigenvector is called {\em PF--eigenvector} of $A$.
189: No other eigenvalue of $A$ has a strictly positive eigenvector.
190: \end{thm}
191:
192: Using this theorem, it is a simple exercise to show the following
193: facts (cf.~\cite{fogg}, Thm.~1.2.7): We consider the
194: {\em relative frequency} of the tiles of type $T_i$ in a tiling $\T$,
195: i.e., the ratio of the number of tiles of type $T_i$ in $\T$ and
196: the number of all tiles in $\T$. Formally, this is defined by
197: $\lim_{r\to \infty} \frac{\# \{ T_i + {\bs x} \, \in \T \, | \, {\bs x}
198: \in r\B^d \}}{\# \{ T+{\bs x} \, \in \T \, | \, {\bs x} \in r\B^d,
199: \, T \in \F \}}$. Since we
200: consider primitive substitution tilings only, it is true that this
201: limit exists. Moreover, this limit exists {\em uniformly}, i.e., for
202: any translate $\T+{\bs x}$ we will obtain the same limit.
203: \begin{prop} \label{freqvol}
204: Let $\sigma$ be a primitive substitution in $\E^d$ with substitution
205: factor $\lambda$ and prototiles $T_1, \ldots, T_m$. Then the
206: PF--eigenvalue of a $A_{\sigma}$ is
207: $\lambda^d$. In particular, a substitution factor is always an algebraic
208: number. \\ The normalized (right) PF--eigenvector ${\bs v}=
209: (v_1,\ldots,v_m)^T$ of $A_{\sigma}$ contains the relative frequencies
210: of the tiles of different types in the tiling in the following sense:
211: The entry $v_i$ is the relative frequency of $T_i$ in $\T$.\\
212: The left PF--eigenvector
213: (resp.~the PF--eigenvector of $A^T$) contains the $d$--dimensional
214: volumes of the different prototiles, up to scaling.
215: \end{prop}
216:
217: Some people don't like the term 'left eigenvector'. Those may
218: replace 'left eigenvector of $A$' by 'eigenvector of $A^T$, written as
219: a row vector' wherever it occurs.
220: In the next section, we will associate with a tiling a related point
221: set: By replacing every prototile $T_i$ with a point in its
222: interior, a tiling $\T$ gives rise to a {\em Delone set} $V_{\T}$.
223: (Note that this defines $V_{\T}$ not uniquely. It depends on the
224: distinct choice of points in the prototiles. But the following is
225: true for any Delone set $V_{\T}$ constructed out of a
226: primitive substitution tiling $\T$ in the described way.)
227: A Delone set is a point set $V \subset \E^d$ which is
228: {\em uniformly discrete} (there exists $r>0$ such that $\# (V \cap
229: r\B^d + {\bs x}) \le 1$ for all ${\bs x} \in \E^d$)
230: and {\em relatively dense} (there exists $R>0$ such
231: that $\# (V \cap R\B^d + {\bs x}) \ge 1$ for all ${\bs x} \in \E^d$).
232: Since the frequencies of the prototiles tiles (cf. remark after Theorem
233: \ref{perron}) are well--defined for all tilings $\T$ considered in this
234: paper, it makes sense to define the density of the associated
235: Delone set $V_{\T}$: $\dens(V_{\T})=\lim_{r\to \infty} \frac{\# V_{\T}
236: \cap r\B^d} {\mu( r \B^d)}$. It is not hard to show the following
237: identity.
238:
239: \begin{prop} \label{densfreqvol}
240: Let $\T$ be a substitution tiling with substitution
241: $\sigma$, let ${\bs v}$ be the right normalized PF--eigenvector of
242: $A_{\sigma}$ (containing the relative frequencies of the prototiles),
243: let ${\bs w}^T$ be the left PF--eigenvector of $A_{\sigma}$, such that
244: ${\bs w}^T$ contains the $d$--dimensional volumes of the prototiles, and
245: $V_{\T}$ a Delone set constructed out of $\T$ by replacing every
246: prototile with a point in its interior. Then
247: \[ \dens(V_{\T}) = \left( {\bs w}^T \cdot{\bs v} \right)^{-1} \]
248: \end{prop}
249:
250: {\em Example:} The golden triangle substitution. The substitution
251: rule, together with substitution factor, substitution matrix etc. is
252: shown below.
253:
254: \begin{figure}[h]
255: \epsfig{file=inflbsp.eps} \hspace{10mm}
256: \begin{minipage}[b]{30mm}
257: $A_{\sigma}:=\begin{pmatrix} 1 & 1 \\ 1 & 0 \end{pmatrix}$
258: \vspace*{5mm}
259: \end{minipage}
260: \caption{'Golden triangles'. A substitution rule in $\E^2$ with
261: two prototiles $T_1,T_2$ and substitution factor $\lambda=\sqrt{\tau}$,
262: where $\tau=\frac{\sqrt{5}+1}{2}$ is the golden mean.
263: \label{inflbsp}}
264: \end{figure}
265: This substitution uses two prototiles $T_1,T_2$.
266: The PF--eigenvalue of the substitution matrix $A_{\sigma}$
267: is $\tau=\lambda^2$. The normalized PF--eigenvector is
268: ${\bs v}:=(\tau^{-1}, \tau^{-2})^T$. Thus, the relative
269: frequency of the tiles of type $T_1$ (resp.~$T_2$) in the tiling is
270: $\tau^{-1}$ (resp.~$\tau^{-2}$). $A_{\sigma}$ is symmetric, thus
271: ${\bs v}^T$ is a left PF--eigenvector. The areas (2--dim volumes) of the
272: two prototiles are $\mu(T_1)=\sqrt{\tau}^3, \mu(T_2)=\sqrt{\tau}$. If
273: we replace every triangle in a corresponding substitution tiling by
274: a point in its interior, then by Prop.~\ref{densfreqvol} the density
275: of the point set is $\left( (\sqrt{\tau}^3,\sqrt{\tau})\cdot(
276: \tau^{-1}, \tau^{-2})^T \right)^{-1} =
277: \frac{\tau^{3/2}}{(\tau^2+1)}=0.56886448\ldots$
278:
279:
280:
281: \subsection{Model Sets} \label{sec:modset}
282: Model sets are a special kind of Delone sets. They show a high degree
283: of local and global order. For example, a model set $V$ and its
284: difference set $V-V := \{ {\bs x} - {\bs y} \, | \, x,y \in V \}$
285: do not differ too much in the following sense (cf.~\cite{mey}): There
286: is a finite set $F$ such that
287: \[ V-V \subseteq V+F. \]
288: Lattices do have this property with $F=\{0\}$, and indeed lattices are
289: special cases of model sets. Unfortunately, there is no reference
290: known to the author which gives a comprehensive overview about model
291: sets. A good reference is \cite{moo}. A comprehensive
292: overview for the case of one--dimensional tilings is contained in
293: \cite{fogg}, mainly chapter 7 and 8. Therein, many connections to
294: combinatorics, ergodic theory and number theory are compiled (but the
295: term 'model set' cannot be found in this book, the term 'geometric
296: representation' is used instead).
297:
298: A model set is defined by a collection of spaces and maps (a
299: so--called cut--and--project scheme) as follows.
300: In (\ref{cps}), let $\Lambda$ be a lattice of full rank in
301: $\E^{d+e}$, $\pi_1,\pi_2$ projections such that
302: $\pi_1 |_{\Lambda}$ is injective, and $\pi_2(\Lambda)$ is dense in
303: $\E^e$. Let $W$ be a nonempty compact set --- the so--called
304: {\em window set} --- with the properties $\cl(\inn(W))=W$
305: and $\mu(\partial W)=0$, where $\mu$ is the Lebesgue measure in $\E^e$.
306: \begin{equation} \label{cps}
307: \begin{array}{ccc}
308: \E^d & \stackrel{\pi_1}{\longleftarrow} \E^{d+e}
309: \stackrel{\pi_2}{\longrightarrow} & \E^e \\
310: \cup & \cup & \cup \\
311: V & \Lambda & W
312: \end{array}
313: \end{equation}
314: Then the set $V= \{ \pi_1({\bs x}) \, | \, {\bs x} \in \Lambda,
315: \pi_2({\bs x}) \in W \}$
316: is called a {\em model set}\footnote{To be precise, this defines a
317: {\em regular} model set. Without the property $\mu(\partial W)=0$,
318: $V$ is still a model set, but not a regular one.}.
319: Obviously, $V$ is a set of points in
320: $\E^d$. A tiling can be obtained by applying certain rules to $V$,
321: e.g., considering the Voronoi cells of $V$. In the case $d=1$ it is
322: quite simple to get a tiling: If $V=\{ {\bs x}_i \, | \, i \in \Z \}$,
323: such that ${\bs x}_i<{\bs x}_{i+1}$, then $\T= \{ [{\bs x}_i, {\bs
324: x}_{i+1} ] \, | \, i \in \Z \}$ is a tiling of $\E^1$.
325:
326: Since $V \subset \pi_1(\Lambda)$ and $\pi_1|_{\Lambda}$ is injective,
327: every ${\bs x} \in V$ corresponds to exactly one point $\pi_1^{-1}({\bs x}) \in
328: \Lambda$, and thus exactly to one point $\pi_2(\pi_1^{-1}({\bs x})) \in W$.
329: Therefore, we are able to define the {\em star map}:
330: \begin{equation} \label{starmap}
331: \ast : \pi_1(\Lambda) \to \E^e, \qquad {\bs x}^{\ast} =
332: \pi_2(\pi_1^{-1}({\bs x}))
333: \end{equation}
334: This allows for a convenient notation. E.g., we have
335: \begin{equation} \label{clvstern}
336: \cl(V^{\ast})=W,
337: \end{equation}
338: where $V^{\ast} = \{ {\bs x}^{\ast} \, | \, {\bs x} \in V \}$.
339:
340: {\em Remark:} In general, $\E^d$ and $\E^e$ in (\ref{cps}) can be
341: replaced by some locally compact abelian groups $G$ and $H$. Then
342: $\mu$ denotes the Haar measure of $H$. This is necessary to show that,
343: for example, the chair tiling is a model set (cf~\cite{bms}).
344: If the substitution matrix $A_{\sigma}$ of a given substitution tiling
345: is {\em unimodular} (i.e., $\det(A_{\sigma} = \pm 1$), then it
346: suffices to consider Euclidean spaces only. If not, then one needs to
347: consider more general groups (cf.~\cite{sing}). In this paper, all
348: occurring substitution matrices are unimodular, therefore we will
349: consider Euclidean spaces only.
350:
351: Given a substitution tiling, one may ask if this tiling can be
352: obtained as a model set $V$. By the following theorem, this is only
353: possible if the substitution factor is a PV number or a Salem number.
354:
355: \begin{defi}
356: Let $\lambda \in \R$ be an algebraic integer. $\lambda$ is called a
357: {\em Pisot Vijayaraghavan number} or shortly {\em PV number},
358: if $|\lambda|>1$, and for all its algebraic conjugates $\lambda_i$
359: holds: $|\lambda_i|<1$. \\
360: If for all conjugates $\lambda_i$ of $\lambda$ holds
361: $|\lambda_i|\le 1$, with equality for at least one $i$, then
362: $\lambda$ is called a {\em Salem number}.
363: \end{defi}
364:
365: \begin{thm}[Meyer] \label{pvsatz}
366: If $V$ is a model set, $1 < \lambda \in \R$ and if $\lambda V \subseteq V$,
367: then $\lambda$ is a PV number or a Salem number.
368: \end{thm}
369:
370: It is known that every substitution $\sigma$ gives rise to at least
371: one tiling $\T$ with the property $\sigma(\T)=\T$. So,
372: if $V$ is a point set derived from $\T$ (where the substitution factor
373: is $\lambda$), say, $V$ being the vertex set of this tiling, then by the
374: definition of a substitution follows $\lambda V \subseteq V$.
375: Now, if $V$ is also a model set, by the theorem above the substitution
376: factor must be a PV number or a Salem number.
377:
378: The remaining things to check in order to show that $V$ is a model set
379: are the following (cf.~(\ref{cps})):
380: \begin{itemize}
381: \item Determine the lattice $\Lambda$,
382: \item determine the window set $W$,
383: \item show $\mu(W)>0$,
384: \item show $\mu(\partial W)=0$,
385: \item show $\dens(V) = \dens(\{ \pi_1({\bs x}) \, |
386: \, {\bs x} \in \Lambda, \pi_2({\bs x}) \in W \})$.
387: \end{itemize}
388: There are standard constructions for obtaining
389: $\Lambda$ and $W$. But the latter three items are hard to decide
390: rigorously in general, or even in special cases. For an example how
391: hard it may be to give a proper proof, cf.~\cite{bs}. In the next section,
392: we will use a duality principle to prove that the three latter items
393: are fulfilled for the considered tiling, thus proving that it arises
394: from a model set (i.e., the associated point set is a model set). But
395: first, we will discuss the above list in more detail.
396:
397: Starting with a substitution tiling, or better a substitution point
398: set $V$ in $\E^1$ (cf.~the remark after (\ref{cps})), the standard
399: construction for the lattice is as follows: Let $\lambda$ be the
400: substitution factor (where $\lambda$ from now on is always assumed to be
401: a PV number) and $\lambda_2,\ldots,\lambda_m$ its algebraic conjugates.
402: Then \[ \Lambda= \langle (1,1,\cdots,1)^T,
403: (\lambda,\lambda_2,\cdots,\lambda_m)^T,
404: (\lambda^2,{\lambda_2}^2,\ldots,{\lambda_m}^2)^T, \ldots,
405: (\lambda^{m-1},{\lambda_2}^{m-1},\ldots,{\lambda_m}^{m-1})^T \rangle_{\Z}. \]
406: Now, $\pi_1$ (resp.~$\pi_2$) can be chosen as the canonical projections
407: from $\E^m \to \E^1$ (resp.~$\E^m \to \E^{m-1}$). To be precise, if
408: ${\bs x}=(x_1,x_2,\ldots,x_m)^T$, then $\pi_1({\bs x})=x_1,
409: \pi_2({\bs x})=(x_2,\ldots, x_m)^T$.
410: At this point, we are already able to compute the window set
411: $W$ numerically: Since the left PF--eigenvector
412: of the substitution matrix contains the the lengths of the
413: prototiles (cf.~remark after Thm.~\ref{perron} above), the
414: lengths can be chosen such that $V \subset \Z[\lambda]$. That is, every
415: ${\bs x} \in V$ can be expressed as ${\bs x}=\alpha_0+\alpha_1 \lambda+
416: \cdots+ \alpha_{m-1} \lambda^{m-1}$, where $\alpha_i \in \Z$.
417: Then
418: \[ {\bs x}^{\ast} = (\alpha_0+\alpha_1 \lambda_2+ \cdots+ \alpha_{m-1}
419: {\lambda_2}^{m-1}, \cdots, \alpha_0+\alpha_1 \lambda_{m-1}+ \cdots+
420: \alpha_{m-1} {\lambda_{m-1}}^{m-1})^T. \]
421: More general, we may apply the star map to $V$, obtaining the
422: (countable) set $V^{\ast}$, which in turn yields the window set
423: $W=\cl(V^{\ast})$. Actually, some of the pictures in the next section
424: were created in essentially in this way: Using the substitution to
425: generate a large finite subset $F \subset V$, where the points are
426: expressed as ${\bs x}$ above, we plot all points of $F^{\ast}$.
427: If this set $F^{\ast}$ appears to be bounded, even if the number of
428: points is increased, then this is an indication that the considered
429: set $V$ is a model set.
430:
431: To prove rigorously that $V$ is actually a model set, one proceeds by
432: showing $\mu(W)>0$ and $\mu(\partial W)=0$. If $W$ is a polyhedron,
433: then this is an easy task. But in the generic case $W$ is of fractal
434: nature, and it may be even hard to prove that $W$ is of positive
435: measure.
436:
437: The last thing to prove is $V=\{ \pi_1({\bs x}) \, | \, {\bs x}
438: \in \Lambda, \pi_2({\bs x}) \in W \}$, and not just a too small
439: subset. More precisely: Let $U:=\{ \pi_1({\bs x}) \, | \, {\bs x}
440: \in\Lambda,\pi_2({\bs x})\in W\}$. Then it must hold
441: $\dens(V)=\dens(U)$. In this context the following
442: theorem turns out to be helpful \cite{sch}.
443:
444: \begin{thm} \label{densvol}
445: Let $V$ be a model set given by (\ref{cps}). Then
446: \[ \dens(V) = \mu(W) / \det \Lambda, \]
447: where $\det \Lambda$ denotes the determinant of the generator matrix of
448: $\Lambda$. Equivalently, $\det \Lambda$ is the volume of any measurable
449: fundamental domain of $\Lambda$.
450: \end{thm}
451:
452:
453:
454: \subsection{Rauzy Fractals} \label{sec:rauzy}
455: As mentioned above, 'Rauzy fractal'
456: is essentially just another term for the window set $W$. The
457: term 'Rauzy fractal' occurs in the context of dynamical systems and
458: number theory, while the terms 'window' or 'window set' arose from the
459: theory of quasicrystals. Originally, 'Rauzy fractal' denoted one special
460: fractal of this kind, namely the one arising from the substitution
461: \[ \sigma(1)=1\;2, \quad \sigma(2) = 1\; 3, \quad \sigma(3)= 1, \]
462: which is analyzed in \cite{r}. Therefore, the term 'generalized Rauzy
463: fractal' is sometimes used to
464: emphasize the fact that one deals with the more general case.
465: (Of course, this is just a symbolic substitution, but it is trivial to
466: formulate the geometric substitution according to Def. \ref{substdef},
467: cf.~comment after (\ref{substn})).
468:
469: According to \cite{fogg}, the definition of a Rauzy fractal is
470: essentially as follows: if $\sigma$ is a substitution for $m$
471: prototiles in $\E^1$, where the factor is a PV number of algebraic
472: degree $m$, then the Rauzy fractal is just the window set $W$ as in
473: (\ref{clvstern}), where one does not care about $\mu(W)>0$ or
474: $\mu(\partial W)=0$ or Theorem \ref{densvol}. Note that therefore it
475: is much more difficult to show that a given substitution point set is
476: a model set than it is to compute the corresponding Rauzy fractal.
477:
478: Before we proceed, let us mention a theorem which is important in this
479: context and which is necessary in the next section.
480:
481: \begin{thm}[\cite{hut}] \label{hutch}
482: Let $(X,d)$ be a complete metric space and $\{f_1,\ldots,f_n\}$ a
483: finite set of contractive maps (i.e., $\exists c<1: \forall x,y \in X:
484: d(f(x),f(y)) \le c\, d(x,y)$). Then there is a unique compact set $K
485: \subset X$ such that $K = \bigcup_{i=1}^n f_i(K)$.
486: \end{thm}
487:
488: This theorem is a consequence of the Banach Fixed Point Theorem, and is
489: easily generalized to partitions of $K$: Let $\{f_1,\ldots,f_n\}$ be
490: as above and $I_j \subseteq \{1,\ldots,n\}$ for $1 \le j \le \ell$,
491: then there is a unique tuple $(K_1,\ldots,K_{\ell})$ of compact sets
492: $K_j$, such that $K_j = \bigcup_{i \in I_j} f_i(K_{j_i})$ for all
493: $1 \le j \le \ell$ (\cite{kli}). In the next section, this theorem will
494: help to show the compactness of the window set $W$.
495:
496:
497:
498: \section{Two Rauzy Fractals and the Duality Principle} \label{sec:zwo}
499: We will describe the duality principle by choosing a certain
500: one--dimensional substitution, constructing the corresponding
501: Rauzy fractal and the associated dual tiling, and then we will
502: use the duality principle to show that the dual tiling is a model set.
503:
504: \subsection{Some Rauzy Fractals} \label{sec:ansatz}
505: It is easy to see, that for any nonnegative integer $n\times n$--matrix
506: $A$ there is a substitution tiling in $\E^1$ with substitution matrix
507: $A$. Actually, from any such matrix one can immediately deduce a
508: substitution, since there are only few geometric restrictions for
509: tilings in one dimension. (The same problem for higher
510: dimensions becomes quite difficult!) Here we consider the following
511: matrices for $n \ge 2$:
512: \[ M_n:=(m_{ij})_{1 \le i,j \le n}, \quad \mbox{where} \; m_{ij} = \left\{
513: \begin{array}{lcl} 1 & : & i+j >n \\ 0 & : & i+j \le n
514: \end{array} \right. \]
515: Obviously all these matrices are primitive.
516: A substitution with substitution matrix $M_n$ for the prototiles
517: $1,2,\ldots,n$ is given by
518: \begin{equation} \label{substn}
519: \begin{array}{ccr} 1 & \to & n \\ 2 & \to & n\!-\!1 \; n \\ & \vdots &
520: \vdots \\ n & \to & 1 \; 2 \cdots n\!-\!1 \; n \end{array}
521: \end{equation}
522: Note that this is just a symbolic substitution. But in one dimension
523: the geometric realization is obtained from this scheme in a unique way:
524: The prototiles are intervals of different lengths, and the lengths are
525: given by the PF--eigenvector of $M_n$. There is a nice trick to deduce the
526: substitution factor directly from this substitutions: For fixed
527: $n \in \Z, n \ge 2$, let $s_k:=\sin(\frac{k \pi}{2n+1})$.
528: Because of the wonderful formula (see \cite{nida})
529: \begin{equation} \label{wformel}
530: \frac{s_k}{s_1} s_i =
531: \sum_{\nu=0}^{k-1}s_{i+1-k+2\nu}
532: \end{equation}
533: it follows $\frac{s_n}{s_1} s_1 = s_n, \frac{s_n}{s_1} s_2 = s_{n-1}+s_n,
534: \ldots, \frac{s_n}{s_1} s_n = s_1 + s_2 + \cdots + s_n$. Therefore the
535: substitution factor (which is the PF--eigenvalue of $M_n$) is
536: $\frac{s_n}{s_1}$, and the length of tile $i$ is $s_i$.
537: Now, we determine which of these substitutions
538: define a Rauzy--fractal. The first condition to be fulfilled is that
539: the substitution factor must be a PV--number\footnote{Or a Salem
540: number, but these don't play a role here, so we don't mention them
541: explicitly. }, cf.~Theorem
542: \ref{pvsatz}. This is the case for $n=2,3,4$ and $7$, and for no
543: other $n \le 30$. (This has been checked by a Maple program, and we
544: don't expect any more PV--numbers as PF--eigenvalues of $M_n$ for
545: larger values of $n$.) The corresponding characteristic polynomials of
546: $M_n$ are:
547: \[ x^2-x-1, \; x^3-2x^2-x+1, \; x^4-2x^3-3x^2+x+1, \;
548: x^7-4x^6-6x^5+10x^4+5x^3-6x^2-x+1. \]
549: The latter two are reducible over \Z. The irreducible polynomials
550: defining $\frac{s_2}{s_1}\, (n=2),\, \frac{s_3}{s_1}\, (n=3), \,
551: \frac{s_4}{s_1} \, (n=4)$ and $\frac{s_7}{s_1} \, (n=7)$ are
552: \begin{equation}\label{charpoly}
553: x^2-x-1, \; x^3-2x^2-x+1, \; x^3-3x^2+1, \; x^4-4x^3-4x^2+x+1.
554: \end{equation}
555: Therefore, the algebraic degree of these numbers is 2,3,3,4, resp.
556: % Note, that all substitution factors (and their powers) are
557: % irrational numbers, thus the resulting tilings are nonperiodic.
558: Here we will focus mainly on the third case ($n=4$). The first case
559: ($n=2$) leads to the well known Fibonacci sequences (see Section
560: \ref{sec:remarks}, or \cite{fogg}, Sections 2.6 and 5.4).
561: The fourth case
562: ($n=7$) will lead to a three-dimensional Rauzy fractal in the end.
563: The third case ($n=4$) will show the methods and the typical problems
564: which may occur, and therefore we discuss this case in the following
565: in detail. This part may be regarded as a recipe how to compute other
566: Rauzy fractals explicitly.
567:
568: \subsection{The case ${\bs n=4}$} \label{sec:n4}
569: In this case we have got four prototiles. Since we consider one--dimensional
570: tilings here, the tiles can be chosen to be intervals. We denote the
571: four prototiles by $L$ ('large'), $M$ ('medium'), $S$ ('small') and
572: $X$ ('extra small'). Their lengths are already known: They are given
573: by a left PV--eigenvector of the substitution matrix, or by
574: (\ref{wformel}), so the values can be chosen as
575: $\frac{s_4}{s_1}, \frac{s_3}{s_1}, \frac{s_2}{s_1},
576: \frac{s_1}{s_1}=1$, resp.~ (or any positive multiples of these
577: values). In order to construct the Rauzy fractal, we need this values
578: as expressions in $\lambda=\frac{s_4}{s_1}$ (cf. Section \ref{sec:modset}).
579: %For this, we use the fact that the
580: %substitution factor is known to be $\lambda=s_4/s_1$.
581: W.l.o.g, we choose $X:=[0,1]$. The substitution now reads
582: \begin{equation}\label{subst}
583: X \to L, \; S \to ML, \; M \to SML, \; L \to XSML.
584: \end{equation}
585: Since $\lambda X = L$ it follows $L=[0,\lambda]$. Let $s,m,\ell$ denote
586: the lengths of $S,M,L$, resp. Since $\ell=\lambda$, we conclude from
587: (\ref{subst}) the equations
588: \[ \begin{array}{rcrcl}
589: \lambda s & = & m+ \lambda & & \\
590: \lambda m & = & s+m+ \lambda & = & s(\lambda+1) \\
591: \lambda^2 & = & 1+s+m+\lambda & = & s(\lambda+1) +1 = \lambda m +1
592: \end{array} \]
593: and therefore $m=\frac{\lambda^2-1}{\lambda}$. From (\ref{charpoly})
594: follows $\lambda^3=3\lambda^2-1$, hence $\lambda^{-1}=3 \lambda -
595: \lambda^2$, and we obtain
596: \[ m = (\lambda^2-1)\cdot(3\lambda-\lambda^2) = \lambda^2 - 2 \lambda. \]
597: Consequently,
598: \[ s = \lambda^2-\lambda-m-1 = \lambda - 1. \]
599: To apply the standard construction of Rauzy fractals, the number of
600: prototiles must equal the algebraic degree of the substitution
601: factor. So we need to get rid of one letter. By (\ref{subst}), a
602: tile $X$ in the tiling is always followed by a tile $S$. Since $x+s=\lambda$,
603: we can merge every pair $XS$ into one tile $L$. It is easy to check
604: that the new substitution
605: \begin{equation} \label{subst2}
606: S \to ML, \; M \to SML, \; L \to LML
607: \end{equation}
608: gives rise to the same tilings as (\ref{subst}) does, up to merging
609: pairs $XS$ into $L$. We identify these tilings with a point set
610: $V$, where $V$ contains all (right) endpoints of the tiles. If $0 \in
611: V$, then $V \subset \Z[\lambda]$, and every point $x \in V$ can be
612: expressed by $x=a + b \lambda + c \lambda^2$ ($a,b,c \in \Z$).
613: The lattice $\Lambda \in \E^3$ now is given by
614: \begin{equation} \label{mlambda}
615: \Lambda = \langle (1,1,1)^T, (\lambda, \lambda_2, \lambda_3)^T, (\lambda^2,
616: \lambda_2^2, \lambda_3^2)^T \rangle_{\Z},
617: \end{equation}
618: where $\lambda_2, \lambda_3$ are the algebraic conjugates of $\lambda$.
619: From $\lambda \lambda_2 \lambda_3 = -1$ and the fact that the values $s_k$
620: occur frequently above, one easily deduces
621: $\lambda_2=-\frac{s_1}{s_2}, \lambda_3=\frac{s_2}{s_4}$.
622: This allows us to write down the star map now (cf.~(\ref{starmap})).
623: Let $0 \in V$. If $x \in V$, then $x= a + b \lambda + c \lambda^2$ for some
624: $a,b,c \in \Z$, and
625: \begin{equation} \label{ast}
626: x^{\ast} = a \begin{pmatrix} 1 \\ 1 \end{pmatrix}
627: + b \begin{pmatrix} \lambda_2 \\ \lambda_3 \end{pmatrix}
628: + c \begin{pmatrix} {\lambda_2}^2 \\ {\lambda_3}^2 \end{pmatrix}.
629: \end{equation}
630: By (\ref{clvstern}), the window set $W$ is just the closure of $V^{\ast}$.
631: A short computation (using $\lambda_i^3= 3 \lambda_i^2-1$ for $i=2,3$)
632: shows
633: \[ (\lambda x)^{\ast} = \begin{pmatrix} \lambda_2 & 0 \\ 0 & \lambda_3
634: \end{pmatrix} x^{\ast} =: Qx^{\ast}. \]
635:
636: In order to show the compactness of $W$, and possibly to draw further
637: conclusions about $W$ (which is the Rauzy fractal wanted), we consider
638: a substitution set $V$ such that
639: $\sigma(V)=V$. (As mentioned above, the considered substitution $\sigma$
640: can be easily formulated for the point set $V$ as well as for the
641: tiling $\T$.) For example, let
642: \begin{equation} \label{sv=v}
643: V = \cdots SMLLM\underline{L}LMLSMLLMLMLSMLLMLLMLSMLL \cdots
644: \end{equation}
645: where the right endpoint of the underlined $L$ is located at zero.
646: (Note, that the condition $\sigma(V)=V$ together with (\ref{sv=v})
647: defines $V$ uniquely.) Denote by $V_L$ ($V_M,V_S$, resp.) the set of
648: all points in $V$ which are right endpoints of intervals $L$ ($M,S$,
649: resp.). Then obviously
650: \[ V=V_L \cup V_M \cup V_S. \]
651: Moreover, from (\ref{subst2}) and $\sigma(V)=V$ follows
652: \begin{equation} \label{vgleich}
653: \begin{array}{lcl}
654: V_L & = & \lambda V \quad \cup \quad \lambda V_L -\lambda^2+\lambda, \\
655: V_M & = & \lambda V -\lambda, \\
656: V_S & = & \lambda V_M - \lambda^2+\lambda.
657: \end{array}
658: \end{equation}
659: This can be seen as follows: One tile of type $M$ and one tile
660: of type $L$ occur in {\em every} substituted tile $T$, not depending
661: on the type of $T$. If $T=[a,b]$, then $\sigma(T)$ contains tiles
662: $[\lambda b - \lambda - m, \lambda b - \lambda]$ (of type $M$) and
663: $[\lambda b -\lambda, \lambda b ]$ (of type $L$). Therefore, from
664: $b \in V$ follows $\lambda b - \lambda \in V_M$,
665: yielding the second equation, and $\lambda b \in V_L$, yielding the first
666: part of the first equation. The other parts of (\ref{vgleich}) are
667: obtained analogously.
668:
669: Applying the star map to (\ref{vgleich}), we obtain the following
670: equations for $V^{\ast}$:
671: \begin{equation} \label{vstern}
672: \begin{array}{lcl}
673: V^{\ast} & = & V_L^{\ast} \quad \cup \quad V_M^{\ast} \quad \cup
674: \quad V_S^{\ast} \\
675: & & \\
676: V_L^{\ast} & = & Q V^{\ast} \quad \cup \quad Q V_L^{\ast}
677: - ({\lambda_2}^2, {\lambda_3}^2)^T + ( \lambda_2 , \lambda_3 )^T \\
678: V_M^{\ast} & = & Q V^{\ast} - ( \lambda_2 , \lambda_3)^T \\
679: V_S^{\ast} & = & Q V_M^{\ast} - ( {\lambda_2}^2 , {\lambda_3}^2 )^T +
680: (\lambda_2 , \lambda_3 )^T
681: \end{array}
682: \end{equation}
683: Obviously, $Q$ defines a contracting linear map, thus all maps
684: appearing in (\ref{vstern}) are contracting. Therefore, Theorem
685: \ref{hutch} applies: There is one unique compact $W=W_L \cup W_M \cup
686: W_S$ which fulfills (\ref{vstern}). This assures the existence of this
687: set, so we obtained a Rauzy fractal $W$.
688:
689: \begin{figure}[h]
690: \epsfig{file=win-kp9.eps} % \hspace{5mm}
691: \epsfig{file=win-sml.eps} \hspace{5mm}
692: \epsfig{file=win-kp7.eps}
693: \caption{Left: The Rauzy fractal $W$ arising from the substitution
694: (\ref{subst2}). Middle: the same set, partitioned according to the
695: distinction of tiles $S$, $M$ and $L$. Right: The Rauzy
696: fractal arising from the substitution (\ref{substn}) for $n=3$: $S
697: \to L, \; M \to ML, \, L \to SML$. \label{wkp9} }
698: \end{figure}
699:
700: To visualize $W$ and its fractal
701: nature, $W$ can be approximated numerically. In fact, we just produce a
702: finite subset of $V$ and apply the star map to it. The resulting
703: points are plotted into a diagram, which is shown in Fig.~\ref{wkp9}
704: (left). If we distinguish the points in the diagram according to their
705: type ($S,M$ or $L$), the result is shown in Fig.~\ref{wkp9} (middle).
706: The set $W$ is the union of the three sets $W_L, W_M, W_S$, where
707: $W_i=\cl(V_i^{\ast})$. From (\ref{vstern}) follows that $W_S$ is
708: an affine image of $W_M$, but not similar to $W_M$, and that $W_M$ is
709: an affine image of $W$.
710:
711:
712:
713: In the same way one can compute the Rauzy fractal for the case $n=3$
714: in (\ref{substn}). The resulting pattern is shown in
715: Fig.~\ref{wkp9}, right. Note, that we did not claim that one of the
716: original substitution sets $V$ is a model set until now. In order to
717: prove this, one needs to prove $\mu(W)>0$, $\mu(\partial W)=0$ and
718: $\dens(V)=\dens(\{ \pi_1({\bs x}) \, | \, {\bs x} \in \Lambda,
719: \pi_2({\bs x}) \in W \})$.
720:
721: \subsection{The dual tiling} \label{sec:dual}
722: Recall that $Q$ in (\ref{vstern}) is a contraction. Therefore $Q':=Q^{-1}$
723: is an expansion. Applying $Q'$ to (\ref{vstern}) and replacing
724: $V_i^{\ast}$ with its closure $W_i$ ($i \in \{S,M,L\}$)
725: yields the following equation system:
726: \begin{equation} \label{dualexp}
727: \begin{array}{lcl}
728: Q' W_L & = & W_L \cup W_M \cup W_S \cup W_L - (\lambda_2,
729: \lambda_3)^T + (1,1)^T \\
730: Q' W_M & = & W_L - (1,1)^T \cup W_M - (1,1)^T
731: \cup W_S - (1,1)^T \\
732: Q' W_S & = & W_M - (\lambda_2, \lambda_3)^T + (1,1)^T
733: \end{array}
734: \end{equation}
735: This defines the substitution $\sigma'$ in $\E^2$ with three prototiles
736: $W_L,W_M,W_S$:
737: \begin{equation} \label{dualsubst}
738: \begin{array}{lcl}
739: \sigma' (\{W_L\}) & = & \{ W_L , W_M , W_S , W_L - (\lambda_2,
740: \lambda_3)^T + (1,1)^T \} \\
741: \sigma' (\{W_M\}) & = & \{ W_L - (1,1)^T , W_M - (1,1)^T
742: , W_S - (1,1)^T \} \\
743: \sigma' (\{W_S\}) & = & \{ W_M - (\lambda_2, \lambda_3)^T + (1,1)^T \}
744: \end{array}
745: \end{equation}
746: In contrary to Def.~\ref{substdef}, the expansion by $Q'$ is not a
747: similarity, but an affinity. Anyway, these substitution rules will
748: yield a tiling of $\E^2$. A part of it is shown in
749: Fig.~\ref{fractilg}. (To be precise, we need to show that there
750: will be no overlaps and no gaps, but this is not essential for the
751: following, since we will consider the corresponding substitution
752: point set, or a modified version where all tiles are polygons.)
753:
754: \begin{figure}[h]
755: \epsfig{file=fractilg.eps} \epsfig{file=polytilg-3.eps}
756: \caption{ \label{fractilg}
757: Left: A patch of the tiling arising from (\ref{dualsubst}). Tiles of
758: type $W_S$ are black, the other tiles come in two slightly different
759: shadings, such that the boundary between tiles of the same type
760: remains visible. Right: A polygonalized version. Tiles and colours
761: correspond to the tiling on the left.}
762: \end{figure}
763:
764: Now, we will prove that the dual tiling is a model set.
765: In analogy to the construction of $V$ out of the original substitution
766: tiling arising from (\ref{subst2}), we choose in every prototile a
767: {\em control point}. A good choice is (cf.~(\ref{vstern})):
768: $c_L:=0 \in W_L$, $c_M:= Qc_L- (\lambda_2,\lambda_3)^T =
769: (\lambda_2,\lambda_3)^T \in W_M$, $c_S:= Qc_M
770: -({\lambda_2}^2,{\lambda_3}^2)^T + (\lambda_2,\lambda_3)^T
771: \in W_S$. It is not too hard to check that the set $\{W_L, W_M
772: -({\lambda_2}^2, {\lambda_3}^2)^T + 2 (\lambda_2,\lambda_3)^T + (1,1)^T\}$
773: gives rise to a point set $V'$ such that $\sigma'(V')=V'$, and $V'$ is
774: a Delone set. To be precise, let $V_0':=\{c_L, c_M-({\lambda_2}^2,
775: {\lambda_3}^2)^T + 2 (\lambda_2,\lambda_3)^T + (1,1)^T\}$, and
776: $\sigma'$ be defined for points as well as for tiles in the canonical way.
777: Then $V_0' \subset \sigma'(V_0')$, and inductively $(\sigma')^k (V_0')
778: \subset (\sigma')^{k+1} (V_0')$. Therefore,
779: \begin{equation} \label{vstrich}
780: V' = \bigcup_{k \ge 1} (\sigma')^k (V_0')
781: \end{equation}
782: is well defined, and it holds $\sigma'(V')=V'$.
783:
784: {\em Remark:} As a first idea, one would try to start with the set
785: $\{ c_L \}$ instead of $V_0'$. In this case, the property
786: $(\sigma')^k(\{ c_L \}) \subset (\sigma')^{k+1}(\{ c_L \})$ is
787: fulfilled, but the resulting point set would be no Delone set, since
788: it possesses arbitrary large holes.
789:
790: By the construction, it holds $V \subset \langle (1,1)^T, (\lambda_2,
791: \lambda_3)^T , ({\lambda_2}^2,{\lambda_3}^2)^T \rangle_{\Z} $. In
792: particular, this shows a
793: curious property of $V'$: If one coordinate of some ${\bs z}=(x,y)^T \in
794: V'$ is known, say, $x=a + b \lambda_2 + c {\lambda_2}^2$ ($a,b,c \in \Z)$,
795: then the other coordinate $y$ is known, too: $y = a + b \lambda_3 + c
796: {\lambda_3}^2$. Another unusual property is that each prototile occurs
797: in just one orientation. This is not true for most known examples of
798: nonperiodic substitution tilings in dimension larger than one.
799:
800: In a similar way as above
801: (cf.~(\ref{vgleich})) we obtain from (\ref{dualsubst}) the following
802: equations, where $V'=V'_S \cup V'_L \cup V'_L$ and $V'_i$ is the set
803: of all points of type $c_i$ in $V'$ ($i \in \{S,M,L\}$).
804: \begin{equation} \label{dualgleich}
805: \begin{array}{lcl}
806: V'_L & = & Q' V'_L \cup Q' V'_M \cup Q' V'_L - (\lambda_2,
807: \lambda_3)^T + (1,1)^T \\
808: V'_M & = & Q' V'_L - (\lambda_2,\lambda_3)^T \cup Q' V'_M - (\lambda_2,
809: \lambda_3)^T \cup Q' V'_S \\
810: V'_S & = & Q' V'_L + (\lambda_2, \lambda_3)^T - 2 ({\lambda_2}^2,
811: {\lambda_3}^2)^T \cup Q' V'_M + (\lambda_2, \lambda_3)^T - 2
812: ({\lambda_2}^2, {\lambda_3}^2)^T
813: \end{array}
814: \end{equation}
815: In order to show that $V'$ is a model set, we have to determine the
816: lattice $\Lambda'$ and the window set $W'$ for $V'$. By the
817: construction of $V'$ follows $\Lambda'=\Lambda$. Since $0 \in V'$, we
818: know that every ${\bs x} \in V'$ is of the form
819: $a(1,1)^T+b(\lambda_2,\lambda_3)^T+c({\lambda_2}^2,{\lambda_3}^2)^T$.
820: Thus our new star map is just the inverse of the original one in
821: (\ref{ast}), namely
822: \[ \star: \pi_2(\Lambda) \to \E^1, \quad {\bs x}^{\star}=
823: \pi_1(\pi_2^{-1}({\bs x}))=a+b \lambda + c \lambda^2. \]
824: Applying $\star$ to (\ref{dualgleich}) yields:
825: \begin{equation} \label{dualglstar}
826: \begin{array}{lcl}
827: {V'_L}^{\star} & = & \lambda^{-1} {V'_L}^{\star} \cup \lambda^{-1}
828: {V'_M}^{\star} \cup \lambda^{-1} {V'_L}^{\star} - \lambda + 1 \\
829: {V'_M}^{\star} & = & \lambda^{-1} {V'_L}^{\star} - \lambda \cup \lambda^{-1}
830: {V'_M}^{\star} - \lambda \cup \lambda^{-1} {V'_S}^{\star} \\
831: {V'_S}^{\star} & = & \lambda^{-1} {V'_L}^{\star} + \lambda - 2 \lambda^2 \cup
832: \lambda^{-1} {V'_M}^{\star} + \lambda - 2 \lambda^2
833: \end{array}
834: \end{equation}
835: Since all occurring maps are contractions, Theorem \ref{hutch} applies,
836: and the unique compact solution is
837: \[ \cl({V_L}^{\star}) = [-\lambda,0] = L, \; \cl({V'_M}^{\star}) =
838: [\lambda-\lambda^2,-\lambda] = M, \; \cl({V'_S}^{\star}) = [-2
839: \lambda^2 + 1, -2 \lambda^2 + \lambda] = S. \]
840: The multiplication of (\ref{dualglstar}) by $\lambda$ yields a
841: distinct geometric realization of our original substitution
842: (\ref{subst2}). So, the dual tiling of the dual tiling is the one we
843: started with. In particular, the window set of $V'$ is $W'=S \cup M
844: \cup L = [-2 \lambda^2 + 1, -2 \lambda^2 + \lambda] \cup
845: [\lambda-\lambda^2,0]$. Therefore $W'$ is compact,
846: \begin{equation} \label{volw}
847: \mu(W')= \lambda^2-1 = 7.290859374\ldots >0
848: \end{equation}
849: and $\mu(\partial W')=0$. All we are
850: left with is to compute $\dens(V')$ and use Theorem \ref{densvol}.
851: If the tiles $W_i$ of the substitution $\sigma'$ in (\ref{dualsubst})
852: would be polygons, this would be an easy task, using the remark after
853: Theorem \ref{perron}. Therefore we replace the fractal tiles
854: $W_S,W_M,W_L$ by appropriate polygons $P_S,P_M,P_L$. Let
855: \[ \begin{array}{rclrcl}
856: a & := & ({\lambda_2}^2 -1 , {\lambda_3}^2 -1 )^T , &
857: b & := & ({\lambda_2}^2, {\lambda_3}^2 )^T, \\
858: c & := & ({\lambda_2}^2 - \lambda_2 , {\lambda_3}^2 -\lambda_3 )^T, &
859: d & := & ({\lambda_2}^2 - \lambda_2 -1, {\lambda_3}^2 -\lambda_3 -1)^T, \\
860: e & := & (2{\lambda_2}^2 - \lambda_2 -1, 2{\lambda_3}^2 -\lambda_3 -1)^T, &
861: f & := & (2{\lambda_2}^2 - 1, 2{\lambda_3}^2 -1)^T, \\
862: g & := & (\lambda_2,\lambda_3)^T, &
863: h & := & ({\lambda_2}^2 + \lambda_2 -1, {\lambda_3}^2 +\lambda_3 -1)^T.
864: \end{array} \]
865:
866: \begin{figure}[h]
867: \epsfig{file=polytiles.eps} \hspace{5mm}
868: \begin{minipage}[b]{34mm}
869: $A_{\sigma'}= \begin{pmatrix} 0 & 1 & 1 \\
870: 1 & 1 & 1 \\ 0 & 1 & 2 \end{pmatrix}$
871: \vspace*{5mm}
872: \end{minipage}
873: \caption{Left: The construction of the polygonal prototiles. Middle:
874: A sketch of the substitution for them. Right: The corresponding
875: substitution matrix $A_{\sigma'}$. Note that $A_{\sigma'} =
876: (A_{\sigma})^T$. \label{polytiles}}
877: \end{figure}
878:
879: Let $\conv({\bs x}_1,\ldots,{\bs x}_n)$ denote the convex hull of
880: ${\bs x}_1,\ldots,{\bs x}_n$. Then (see Fig.~\ref{polytiles})
881: \begin{equation}
882: \begin{split}
883: P_S & := \conv(a-b,a,h,h-b), \\
884: P_M & := \conv(e-g,e,d,d-g) \, \cup \, \conv(c-g,c,e,e-g), \\
885: P_L & := \conv(0,e,c,b) \, \cup \, \conv(0,g,h,d,e)
886: \end{split}
887: \end{equation}
888: It is quite simple (but lengthy) to show that by replacing $W_i$ with
889: $P_i$ in (\ref{dualsubst}), the substitution $\sigma'$ leads to a tiling,
890: without any gaps or overlaps.
891: (Essentially, one uses the definition of a substitution tiling in
892: Definition \ref{substdef} and the primitivity of the substitution
893: matrix. To show that there are no gaps and no
894: overlaps it suffices to consider all vertex configurations in the
895: first $n$ substitutions $\sigma'(T),(\sigma')^2(T), \ldots,
896: (\sigma')^n(T)$ for appropriate $n$ and some prototile $T$. So, in any
897: certain case this is a finite
898: problem. Here $n=4$ and $T=P_L$ will do.) It is also quite simple (but
899: lengthy again) to compute the areas of the prototiles
900: $P_S,P_M,P_L$. Unfortunately, we cannot use the fact that the left
901: eigenvector of the corresponding substitution matrix reflects the
902: ratio of the areas, since the substitution for the polygonal tiles
903: does not preserve the shape any longer. E.g., $\sigma'(W_S)$ is congruent
904: to $Q'W_S$, but $\sigma'(P_S)$ is {\em not} congruent to $Q'P_S$.
905: Altogether, we obtain
906: \begin{equation}
907: \begin{split}
908: \mu(P_S) & = |\lambda_2| \lambda_3 (\lambda_3-\lambda_2) = \frac{s_1}{s_4}
909: \left( \frac{s_2}{s_4} + \frac{s_1}{s_2} \right), \\
910: \mu(P_M) & = \lambda_3-\lambda_2 = \frac{s_2}{s_4} + \frac{s_1}{s_2}, \\
911: \mu(P_L) & = {\lambda_2}^2 - {\lambda_3}^2 + 2 |\lambda_2| + 2 \lambda_3 -
912: |\lambda_2| {\lambda_3}^2 - \lambda_3 {\lambda_2}^2 = 1 + \frac{s_4
913: s_1}{{s_2}^2}, \\
914: \end{split}
915: \end{equation}
916: where the last equality requires an excessive use of (\ref{wformel}).
917: The vector $(\mu(P_S),\mu(P_M),\mu(P_L))$ indeed is a left
918: eigenvector of the substitution matrix $A$ (cf.~Fig.~\ref{polytiles})
919: corresponding to the PF eigenvalue $\lambda$. A right eigenvector
920: corresponding to $\lambda$ is $(s_2,s_3,s_4)^T$. Therefore,
921: the normalized right eigenvector of $A$ --- which entries, by Theorem
922: \ref{freqvol}, are the
923: relative frequencies $p_i$ of tile types $P_i$ ($i \in \{S,M,L\}$) in
924: the tiling --- is $(p_S,p_M,p_L)^T := \frac{1}{s_2 + s_3 + s_4}
925: (s_2,s_3,s_4)$. Since every point of $V'$ corresponds to exactly one
926: tile in the polygonal tiling, by Prop. \ref{densfreqvol} the relative
927: density of $V'$ is
928: \begin{equation} \label{densv}
929: \begin{split}
930: \dens(V') & =
931: \left( (\mu(P_S),\mu(P_M),\mu(P_L))\cdot(p_S,p_M,p_L)^T \right)^{-1} \\
932: & = (s_2+s_3+s_4) \left( \frac{s_1{s_2}^2}{s_4^2} + \frac{s_1^2}{s_4}
933: + \frac{s_2 s_3}{s_4}+\frac{s_1 s_3}{s_2}+ s_4
934: + \frac{s_1 {s_4}^2}{{s_2}^2} \right)^{-1} \\
935: & = 0.8100954858 \ldots
936: \end{split}
937: \end{equation}
938: In order to apply Theorem \ref{densvol} we need to compute $\det
939: \Lambda$ (cf.~(\ref{mlambda})). A rather lengthy computation (or the
940: usage of a computer algebra system, here: Maple) shows
941: $\det \Lambda = \frac{27}{64 {s_1}^2 {s_2}^2 {s_4}^2}$. Using
942: elementary trigonometric formulas yields
943: \[
944: \begin{split}
945: s_1 s_2 s_4 & = \frac{1}{4} (s_3+s_4 - s_2 - s_1 ) \\
946: & = \frac{1}{4} \left( s_3 + 2 \cos \left( \frac{3 \pi}{9} \right)
947: s_1 -s_1 \right) = \frac{1}{4} s_3 = \frac{\sqrt{3}}{8},
948: \end{split}
949: \]
950: and therefore $\det \Lambda = 9$. Putting together this with
951: (\ref{volw}) and (\ref{densv}) in view of Theorem \ref{densvol} we
952: obtain \[ \mu(W')/\dens(V') = 9 =\det(\Lambda). \]
953: The leftmost equality requires again a very lengthy computation, or
954: the usage of Maple. This completes the proof, and we obtain the
955: following theorem.
956: \begin{thm} \label{dersatz}
957: The set $V'$ in (\ref{vstrich}), arising from the substitution $\sigma'$
958: in (\ref{dualsubst}), is a model set.
959: \end{thm}
960: One may ask why we do not state that the original set $V$ arising from
961: the substitution (\ref{subst2}) is a model set. The reason is that
962: there is one gap, namely, the value of $\mu(W)$. For example, one needs
963: to show that the tiling by the fractal tiles $W_S,W_M,W_L$ is not
964: overlapping. This seems to be clear from the pictures, but we did not
965: give a rigorous proof here. Indeed, this will require some more work,
966: and we refer to further publications. Anyway, there are good reasons
967: to assume that the areas of the
968: fractal tiles $W_S,W_M,W_L$ are equal to the areas of the polygonal
969: tiles $P_S,P_M,P_L$, resp. Proving this will complete the
970: proof that the original set $V$ is a model set, too.
971:
972: \section{Further Remarks} \label{sec:remarks}
973: The question whether a given structure is a model set is motivated for
974: example by the question for the diffraction property of this structure.
975: It is known that model sets have a pure point diffraction image,
976: without any continuous part. For details, cf.~\cite{bmrs}.
977:
978: We did not emphasize it throughout the text, but all tilings considered
979: here are nonperiodic. For the one--dimensional tilings this is a
980: consequence of the fact that the relative frequencies of the tiles are
981: irrational numbers. If a one--dimensional tiling is periodic, then the
982: relative frequencies are rational numbers. Anyway, all occurring
983: two--dimensional tilings are nonperiodic, too. This can be shown with
984: standard tools from the theory of nonperiodic substitution tilings
985: (cf.~statement 10.1.1 in \cite{gs}).
986:
987: An important point in proving Theorem \ref{dersatz} is that we know
988: the occurring values in terms of trigonometric expressions, thus
989: it is possible to make exact computations of $\mu(W')$, $\dens(V')$,
990: $\det(\Lambda)$ etc. In general, the fractal nature of the occurring
991: window sets may forbid this. But if all occurring sets are polytopes
992: the things become easier. In the following, some examples are listed
993: where the duality principle provides short and simple proofs that
994: these considered substitution tilings are model sets.
995:
996: {\bf Fibonacci sequences:} The symbolic substitution
997: $S \to L,\, L \to LS$ gives rise to the widely examined Fibonacci
998: sequences. (Not to be confused with the Fibonacci numbers
999: $1$,$1$,$2$,$3$,$5$,$8$,$13$,$\ldots$, sometimes also called Fibonacci
1000: sequence. If one considers the first substitutions of $S$, then the
1001: relation becomes clear: $L$, $LS$, $LLS$, $LSLLS$, $LSLLSLSL$, $\ldots$).
1002:
1003: A geometric representation as a substitution tiling
1004: (Fibonacci tiling) is for
1005: example given by $L:=[0,1],S=[1, \tau]$ ($\tau$ the golden ratio)
1006: and \[ \sigma (\{S\}) = \{ L+\tau \}, \; \sigma( \{L\} ) = \{L,S\}. \]
1007: By the standard construction (cf.~Section \ref{sec:modset}), we obtain
1008: for some $V$ which fulfills $\sigma(V)=V$ (namely, $V= \bigcup_{k \ge
1009: 0} \sigma^{2k}(\{ L, L-1\})$, up to one point this will do) the
1010: equations
1011: \[ \begin{split}
1012: V_L^{\ast} & = -\tau^{-1} V_L^{\ast} \cup -\tau^{-1} V_S^{\ast} \\
1013: V_S^{\ast} & = -\tau^{-1} V_L^{\ast} + 1
1014: \end{split}, \]
1015: to be read as equations in $\E^1$. The unique compact solution is
1016: \[ \cl(V_L^{\ast}) = W_L = [-1, \tau^{-1}] , \;
1017: \cl(V_S^{\ast}) = W_S = [\tau^{-1}, \tau]. \]
1018: The dual tiling is given by the following substitution:
1019: \[ \begin{split}
1020: \sigma'(\{W_L\}) & = \{W_L,W_S\} \\
1021: \sigma'(\{W_S\}) & = \{W_L - \tau \}
1022: \end{split} \]
1023: Note that the substitution factor is $-\tau$. Nevertheless, the
1024: resulting family of substitution tilings for $\sigma'$ is equal to
1025: these for $\sigma$. That means the dual tilings of Fibonacci tilings
1026: are Fibonacci tilings again; or shortly: Fibonacci tilings are
1027: self--dual. Moreover, a short computation yields that the equation in
1028: Theorem \ref{densvol} is fulfilled, and therefore the Fibonacci
1029: sequences are model sets.
1030:
1031: \begin{figure}[h]
1032: \epsfig{file=ammch.eps}
1033: \caption{\label{ammch} Left: The substitution for the Ammann
1034: chair. Right: A part of a corresponding substitution tiling.}
1035: \end{figure}
1036:
1037: {\bf Golden triangles and Ammann chair:}
1038: The substitution for the golden triangles (cf.~Fig.~\ref{inflbsp})
1039: defines a model set, if one replaces every triangle with the vertex
1040: at its right angle. This was essentially shown in \cite{dvo} in
1041: detail\footnote{The authors used {\em all} vertices in the tiling
1042: instead of one vertex of each triangle, but their result is easily
1043: transferred to the latter case.}. It turns out that the golden
1044: triangle tilings are dual to the so--called Ammann chair tilings (or
1045: 'A2 tilings', cf.~\cite{gs}). A sketch of the substitution of the
1046: latter is shown in Fig.~\ref{ammch}, together with a part of the
1047: tiling. The substitution factor and the substitution matrix are equal
1048: to those of the golden triangle substitution. All occurring tiles, and
1049: all occurring window sets are polygons. Therefore, it is simple to
1050: compute the areas and the densities, thus proving --- with the help of
1051: Theorem \ref{densvol} --- that both substitution tilings (resp.~the
1052: corresponding point sets) are model sets.
1053: % In the golden triangle tilings, each of the two prototiles
1054: % do occur in four different orientations. Therefore we distinguish
1055: % eight prototiles, four small ones and four large ones, according to
1056: % their orientation (cf.~Fig.~\ref{ammch}, left).
1057:
1058: The duality of the Ammann chair tilings and the golden triangle
1059: tilings was already indicated in \cite{gel} briefly. But a proof of
1060: this needs some care. For example, one has to pay attention to the
1061: fact, that each prototile occurs in four different orientations, not
1062: just in one, as it is true for the tilings in Fig.~\ref{fractilg}.
1063:
1064: {\bf Variants of the tilings in Section \ref{sec:zwo}:}
1065: The tilings considered in the last section are a rich source of other
1066: substitutions, or other Rauzy fractals. If we restrict to the case
1067: $n=4$, then the underlying lattice $\Lambda$ will always be the same
1068: (namely, as in (\ref{mlambda})). One possible modification is to
1069: permute the letters in the substitution. E.g., the substitution
1070: \begin{equation} \label{substperm}
1071: S \to ML, \; M \to MLS, \; L \to LML
1072: \end{equation}
1073: instead of (\ref{subst2}) yields the Rauzy fractal shown in
1074: Fig.~\ref{mehrrosi} (left). Another variant is to use the
1075: matrix $A_{\sigma'}$ of Fig.~\ref{polytiles} (the transpose of the
1076: original substitution matrix $A_{\sigma}$) as a substitution matrix
1077: for a one--dimensional substitution with substitution factor
1078: $\lambda$. One possibility is the substitution
1079: \begin{equation} \label{substtransp}
1080: S \to M, \; M \to SML, \; L \to SMLL,
1081: \end{equation}
1082: where $S,M,L$ are intervals of length $\frac{s_1 s_2}{s_4}, s_2,s_4$,
1083: resp. This substitution yields the Rauzy fractal shown in
1084: Fig.~\ref{mehrrosi} (right). Note that in this case the set $W_M$ is not
1085: connected, it consists of two parts which are similar to each other.
1086: The substitution tiling arising from this Rauzy fractal (resp.~from
1087: the corresponding equation system, multiplied by $Q'$,
1088: cf.~(\ref{dualexp})) would possess a prototile, namely, $W_M$, which is
1089: disconnected. Note, that we did not exclude disconnected
1090: (proto--)tiles in our definition of (substitution) tilings.
1091: % Moreover, there seems to exist a point ${\bs x}$ in $W_S$
1092: % with the property, that $W_S \setminus \{ {\bs x} \}$ is not
1093: % connected. If so, then the analogous statement would hold for the
1094: % two parts of $W_M$, resp.
1095:
1096: \begin{figure}[h]
1097: \epsfig{file=wkp9-alt1.eps} \hfill \epsfig{file=wkp9-alt2.eps}
1098: \caption{\label{mehrrosi}
1099: Two more Rauzy fractals, the left one arising from the substitution
1100: (\ref{substperm}), the right one arising from the substitution
1101: (\ref{substtransp}) }
1102: \end{figure}
1103:
1104: Combining these two modifications leads to even more examples, all of
1105: them based on the lattice $\Lambda$. All the two--dimensional tilings
1106: are model sets, which is easily proven by using the results of Section
1107: \ref{sec:zwo}.
1108: From the case $n=3$ one can also construct different tilings by
1109: permuting the letters. In contrary, all variants of the case $n=2$
1110: will yield Fibonacci sequences. Anyway, in the cases $n=3, n=4$ it may
1111: be interesting to examine the different possibilities with respect to
1112: common properties of the corresponding Rauzy fractals.
1113:
1114:
1115: \begin{thebibliography}{ZZZZZ}
1116:
1117: \bibitem[{\sc BMRS}]{bmrs}
1118: M.~Baake, R.V.~Moody, C.~Richard, B.~Sing:
1119: Which distribution of matter diffracts?
1120: \textit{Quasicrystals: Structure and Physical Properties},
1121: ed.~Trebin, Hans-Rainer, Wiley-VCH, Berlin (2003) pp 188--208
1122: \texttt{math-ph/0301019}
1123:
1124: \bibitem[{\sc BMSc}]{bms}
1125: M.~Baake, R.V.~Moody, M.~Schlottmann:
1126: Limit--(quasi)periodic point sets as quasicrystals with $p$-adic
1127: internal spaces, {\it J.Phys. {\bf A}: Math. Gen.} {\bf 31} (1998)
1128: 5755--5765 \texttt{math-ph/9901008}
1129:
1130: \bibitem[{\sc BS}]{bs}
1131: M.~Baake, B.~Sing:
1132: Kolakoski-(3,1) is a (deformed) model set,
1133: \textit{Can.~Math.~Bulletin} \textbf{47} (2004) 168--190
1134: \texttt{math.MG/0206098}
1135:
1136: \bibitem[{\sc DvO}]{dvo}
1137: L.~Danzer, G.~van Ophuysen:
1138: A species of planar triangular tilings with inflation factor
1139: $\sqrt{-\tau}$,
1140: {\em Panjab Univ.~Res.~Bull.} {\bf 50} (2000) 137--175
1141:
1142: \bibitem[{\sc Gel}]{gel}
1143: G.~Gelbrich:
1144: Fractal Penrose tiles II: Tiles with fractal boundary as duals of
1145: Penrose triangles,
1146: \textit{Aequationes Math.} \textbf{54} (1997) 108--116
1147:
1148: \bibitem[{\sc GSh}]{gs}
1149: B.~Gr\"unbaum, G.C.~Shephard:
1150: {\it Tilings and Patterns},
1151: Freeman, New York (1987)
1152:
1153: \bibitem[{\sc Hut}]{hut}
1154: J.E.~Hutchinson:
1155: Fractals and self--similarity,
1156: \textit{Indiana Univ.~Math.~J.} \textbf{30} (1981) 713--747
1157:
1158: \bibitem[{\sc Kli}]{kli}
1159: R.~Klitzing:
1160: \textit{Reskalierungssymmetrien quasiperiodischer Strukturen},
1161: Ph.D.~thesis, Univ.~T\"ubingen,
1162: Dr.~Kova\u{c}, Hamburg (1996)
1163:
1164: \bibitem[{\sc Mey}]{mey}
1165: Y.~Meyer:
1166: \textit{Algebraic numbers and harmonic analysis},
1167: North--Holland Math. Lib. {\bf 2} North--Holland, Amsterdam
1168: (1972)
1169:
1170: \bibitem[{\sc Moo}]{moo}
1171: R.V.~Moody:
1172: \textit{Meyer sets and their duals},
1173: The mathematics of long-range aperiodic order.
1174: Proceedings of the NATO Advanced Study Institute (Waterloo, Ontario,
1175: 1995) NATO ASI Ser., Ser. C, Math. Phys. Sci. 489,
1176: Kluwer, Dordrecht (1997) pp 403--441
1177:
1178: \bibitem[{\sc ND}]{nida}
1179: K.--P.~Nischke, L.~Danzer:
1180: A construction of inflation rules based on n--fold symmetry,
1181: \textit{Discr.~Comp.~Geom.} \textbf{15} (1996) 221--236
1182:
1183: \bibitem[\textsc{PyF}]{fogg}
1184: N.~Pytheas Fogg:
1185: Substitutions in Dynamics, Arithmetics and Combinatorics,
1186: \textit{Lecture Notes in Mathematics} \textbf{1794},
1187: Springer, Berlin (2002)
1188:
1189: \bibitem[\textsc{R}]{r}
1190: G.~Rauzy:
1191: Nombres alg\'ebriques et substitutions.
1192: Bull.~Soc.~Math.~Fr.~\textbf{110} (1982) 147--178
1193:
1194: \bibitem[\textsc{Sc}]{sch}
1195: M.~Schlottmann:
1196: Cut--and--project sets in locally compact abelian groups,
1197: \textit{Quasicrystals and Discrete Geometry},
1198: ed.~J.~Patera, Fields Institute Monographs \textbf{10},
1199: AMS, Providence, RI (1998) pp 247--264
1200:
1201: \bibitem[\textsc{S}]{sing}
1202: B.~Sing:
1203: Pisot substitutions and (limit-) quasiperiodic model sets,
1204: Talk at: MASCOS Workshop on Algebraic Dynamics,
1205: UNSW, Sydney, Australia (Feb. 2005) \\
1206: \texttt{http://www.math.uni-bielefeld.de/baake/algdyn/sing.pdf}
1207:
1208:
1209: \end{thebibliography}
1210:
1211: \end{document}
1212:
1213:
1214:
1215: