1: \documentclass[11pt]{amsart}

2: \usepackage{amsmath}

3: \usepackage{amssymb,bbm}

4: \usepackage{epsfig}

5: \usepackage{a4wide}

6: \usepackage{pifont}

7: 

8: \newcommand{\N}{\ensuremath{\mathbb{N}^{}_0}}

9: \newcommand{\Z}{\ensuremath{\mathbb{Z}}}

10: \newcommand{\R}{\ensuremath{\mathbb{R}}}

11: \newcommand{\Q}{\ensuremath{\mathbb{Q}}}

12: \newcommand{\C}{\ensuremath{\mathbb{C}}}

13: \newcommand{\Sb}{\ensuremath{\mathbb{S}}}

14: \newcommand{\X}{\ensuremath{\mathbb{X}}}

15: \newcommand{\Sc}{\ensuremath{{\mathcal S}}}

16: \newcommand{\D}{\ensuremath{{\mathcal D}}}

17: \newcommand{\F}{\ensuremath{{\mathcal F}}}

18: \newcommand{\gL}{\varLambda}

19: \newcommand{\gG}{\varGamma}

20: \newcommand{\gT}{\varTheta}

21: \newcommand{\smc}{\scriptscriptstyle\square}

22: 

23: \newcommand{\todo}[1]{[\textnormal{\ding{46}\ding{55}} #1]}

24: 

25: \DeclareMathOperator{\dd}{d\!}

26: \DeclareMathOperator{\dens}{dens}

27: \DeclareMathOperator{\vol}{vol}

28: 

29: \newtheorem{thm}{Theorem}

30: \newtheorem{prop}[thm]{Proposition}

31: \newtheorem{lem}[thm]{Lemma}

32: \newtheorem{cor}[thm]{Corollary}

33: 

34: \newtheorem{claim}{Claim}

35: \newtheorem{conj}{Observation}

36: 

37: \theoremstyle{definition}

38: 

39: \renewcommand{\baselinestretch}{1.06}

40: 

41: \begin{document}

42: 

43: \title[A radial analogue of Poisson's summation formula]{A radial

44:   analogue of Poisson's summation formula\\[2mm] with applications to

45:   powder diffraction\\[2mm] and pinwheel patterns}  

46: 

47: \author{Michael Baake}

48: \address{Fakult\"at f\"ur Mathematik, Universit\"at Bielefeld, 

49: Postfach 100131, 33501 Bielefeld,  Germany}

50: \email{\texttt{mbaake@math.uni-bielefeld.de}, 

51: \texttt{dirk.frettloeh@math.uni-bielefeld.de}}

52: \urladdr{\texttt{http://www.math.uni-bielefeld.de/baake},

53: \texttt{http://www.math.uni-bielefeld.de/baake/frettloe/}}

54: 

55: \author{Dirk Frettl\"oh}

56: % \address{Fakult\"at f\"ur Mathematik, Universit\"at

57: %   Bielefeld, 33501 Bielefeld, Germany}

58: % \email{\texttt{dirk.frettloeh@math.uni-bielefeld.de}}

59: % \urladdr{\texttt{http://www.math.uni-bielefeld.de/baake/frettloe/}}

60: 

61: \author{Uwe Grimm}

62: \address{

63: Department of Mathematics, The Open University, Walton Hall,

64: Milton Keynes MK7 6AA, UK}

65: \email{u.g.grimm@open.ac.uk}

66: \urladdr{http://mcs.open.ac.uk/ugg2/}

67: 

68: 

69: \begin{abstract} 

70: Diffraction images with continuous rotation symmetry arise from

71: amorphous systems, but also from regular crystals when investigated

72: by powder diffraction. On the theoretical side, pinwheel patterns and

73: their higher dimensional generalisations display such symmetries as

74: well, in spite of being perfectly ordered. We present first steps and

75: results towards a general frame to investigate such systems, with

76: emphasis on statistical properties that are helpful to understand and

77: compare the diffraction images. An alternative substitution rule for

78: the pinwheel tiling, with two different prototiles, permits the

79: derivation of several combinatorial and spectral properties of this

80: still somewhat enigmatic example. These results are compared with 

81: properties of the square lattice and its powder diffraction.

82: 

83: \end{abstract} 

84: 

85: \maketitle

86: 

87: \section{Introduction} \label{seq:intro}

88: 

89: Since the discovery of quasicrystals some 20 years ago, mathematicians

90: and physicists have gained a reasonable understanding of aperiodically

91: ordered systems, in particular of those obtained from the projection

92: method. Such sets are called cut and project sets, or model sets

93: \cite{moo}. These are Delone sets of finite local complexity with

94: respect to translations, which also means that any finite patch occurs

95: in finitely many orientations only.

96: 

97: Much less is known about aperiodically ordered systems with local

98: patches occurring in infinitely many orientations, such as the

99: pinwheel tiling of the plane or its three-dimensional counterpart,

100: see \cite{rad} and references therein. Arguably, these are closer to

101: amorphous systems, but still perfectly ordered. To our knowledge, no

102: mathematically satisfactory frame for the analysis of radially

103: symmetric systems and the comparison of their spectral properties has

104: been developed so far. It is the aim of this contribution to show

105: first steps in this direction, by combining results and methods from

106: discrete geometry with the more recent approach of mathematical

107: diffraction theory.

108: 

109: Our guiding examples are the square lattice and the pinwheel tiling of

110: the plane. The diffraction of the square lattice is well understood,

111: and it is not difficult to get some insight into its powder

112: diffraction. The latter emerges from the presence of many grains in

113: random mutual orientations, and thus requires a setting with circular

114: symmetry.

115: 

116: Circular symmetry is also a fundamental property of the pinwheel

117: tiling, resp.\ its compact hull. This tiling has recently been

118: reinvestigated from the autocorrelation and diffraction point of view

119: \cite{mps}. However, hardly any explicit calculation exists in the

120: literature, the reason being the enigmatic nature of the substitution

121: generated pinwheel tiling.  New insight is gained by means of an

122: alternative construction on the basis of a substitution rule with two

123: distinct prototiles. This gives access to quantities such as

124: frequencies (hence also to the frequency module), distance sets and

125: the ring structure of the diffraction measure.

126: 

127: In Section~\ref{sec:rpsf}, we derive a radial analogue of Poisson's

128: summation formula for tempered distributions, which is needed in the

129: following analysis. In the last section, we outline the general

130: structure by presenting a number of results. They are clearly

131: distinguished according to their present status into theorems (with

132: proofs or references), claims (with a sketch of the idea and a

133: reference to future work) and observations (based on numerical or

134: preliminary evidence). Since our alternative substitution will no

135: doubt enable other developments as well, we hope that further progress

136: is stimulated by the results presented in this paper.

137: 

138: \subsection{Notation and preliminaries}

139: 

140: A rotation through $\alpha$ about the origin is denoted by

141: $R^{}_{\alpha}$, and $B^{}_r(x)$ is the closed ball of radius $r$

142: centred in $x$.  Whenever we speak of an \emph{absolute frequency}

143: (e.g., of a point set), we mean the average number per unit volume,

144: whereas \emph{relative frequency} of a subset of objects is used with

145: respect to the entire number of objects. The absolute frequency of the

146: points in a point set $X$ (if it exists) is also called the

147: \emph{density} of $X$, denoted by $\dens(X)$. The set of non-negative

148: integers is called $\N$, and the unit circle is

149: $\Sb^{1}=\{z\in\C:|z|=1\}$.  If $M$ is a locally finite point set, the

150: corresponding \emph{Dirac comb} $\delta^{}_M := \sum_{x \in M}

151: \delta^{}_x$, where $\delta^{}_x$ is the normalised point (or Dirac)

152: measure in $x$, is a well-defined measure. The Fourier transform of

153: $g$ is denoted by $\widehat{g}$. Whenever we use Fourier transform for

154: measures below, we are working in the framework of tempered

155: distributions \cite{sch}.

156: 

157: \section{A radial analogue of Poisson's summation formula}

158: \label{sec:rpsf}

159: 

160: The diffraction pattern of an ideal crystal, supported on a point

161: lattice $\gG \subset \R^d$, can be obtained by the Poisson

162: summation formula 

163: (PSF) for lattice measures or Dirac combs \cite{sch,cor1,cor2} 

164: \begin{equation} \label{eq:psf}

165: \widehat{\delta}^{}_{\gG} = \dens(\gG) \cdot

166: \delta^{}_{\gG^{\ast}},

167: \end{equation}

168: where $\gG^{\ast} := \{ x \in \R^d : x \!\cdot\! y \in \Z$ for all $y

169: \in \gG \}$ is the dual lattice. The distribution-valued (or

170: measure-valued) version follows from the classical PSF, compare

171: \cite{cor1,ik}, via applying it to a (compactly supported) Schwartz

172: function. It is often used to derive the diffraction measure

173: $\widehat{\gamma}^{}_{\omega}$ of a lattice periodic measure $\omega =

174: \varrho \ast \delta^{}_{\gG}$ (with $\varrho$ a finite measure). One

175: obtains the autocorrelation measure

176: \begin{equation}  \label{eq:autocor} 

177: \gamma^{}_{\omega} = \dens(\gG) \cdot (\varrho \ast

178: \widetilde{\varrho}\, ) \ast \delta^{}_{\gG} ,

179: \end{equation}

180: where $\widetilde{\varrho}(g) := \overline{\varrho(\widetilde{g})}$

181: with complex conjugation $\overline{\vphantom{g}.}$ and

182: $\widetilde{g}(x)=\overline{g(-x)}$, and the diffraction measure

183: \[

184: \widehat{\gamma}^{}_{\omega} = ( \dens(\gG))^2 \cdot |

185: \widehat{\varrho} \, |^2 \cdot \delta^{}_{\gG^{\ast}} ,

186: \] 

187: see \cite{hof,baa} for details. Our interest is to extend this

188: approach to situations with circular (or spherical) symmetry. 

189:  

190: Let us first consider the square lattice $\Z^2$, and recall that its

191: circular shells have radii precisely in the set 

192: \begin{equation}\label{defrad}

193:  \D^{}_{\smc} := \{ r \ge 0 : r^2 = m^2

194: + n^2 \; \mbox{with $m,n \in \Z$} \} =

195: \{0,1,\sqrt{2},2,\sqrt{5},2 \sqrt{2}, 3, \ldots \} . 

196: \end{equation}

197: This is the set of non-negative numbers whose squares are integers

198: that contain primes $p \equiv 3 \mod 4$ only to even

199: powers. Moreover, on a shell of radius 

200: $r \in \D^{}_{\smc}$, one 

201: finds finitely many lattice points, their number being given by 

202: \begin{equation} \label{etaq}

203:  \eta^{}_{\smc} (r) =   

204: \begin{cases} 

205: 1, & r=0, \\

206: 4 a(r^2), & r \in \D^{}_{\smc}\setminus\{0\}. 

207: \end{cases}

208: \end{equation}

209: Here, $a(n)$ is the number of ideals of norm $n$ in $\Z[i]$, the ring

210: of Gaussian integers. This is a multiplicative arithmetic function,

211: thus specified completely by its values at prime powers (see

212: \cite{bg,haw}).  They are given by

213: \[ 

214: a(p^{\ell}) = \begin{cases} 1, & \text{if $p=2$}, \\

215: \ell+1, &  \text{if $p \equiv 1\bmod 4$}, \\ 

216: 0, &  \text{if $p\equiv 3 \bmod 4$ and $\ell$ odd}, \\ 

217: 1, &  \text{if $p\equiv 3 \bmod 4$ and $\ell$ even}. \end{cases}

218: \] 

219: As $\Z^2$ is self-dual as a lattice, with $\dens(\Z^2)=1$, the PSF

220: \eqref{eq:psf} simplifies to $\widehat{\delta}^{}_{\Z^2} =

221: \delta^{}_{\Z^2}$.

222: 

223: Choose an irrational number $\alpha \in (0,1)$. By Weyl's lemma

224: \cite{kn}, the sequence $(n \alpha \bmod 1)_{n \ge 1}$ is uniformly

225: distributed in $(0,1)$. Consider the sequence 

226: $(\omega^{}_N)^{}_{N \ge  1}$ of measures defined by

227: \begin{equation} \label{eq:sumrg} 

228:  \omega^{}_N = \frac{1}{N} \sum_{n=1}^N \delta^{}_{R^n \Z^2}, 

229: \end{equation}

230: where $R=R^{}_{2 \pi \alpha}$ is the rotation through $2 \pi \alpha$.

231: If $|x|=r>0$, the sequence $(R^n x)^{}_{n \ge 1}$ is uniformly

232: distributed on $\partial B^{}_r(0)$, again by Weyl's lemma. Observe

233: that all lattices $R^n \Z^2$ share the same set of possible shell

234: radii, namely $\D^{}_{\smc}$ of \eqref{defrad}.  This implies that,

235: given an arbitrary compactly supported continuous function $\varphi$,

236: one has

237: \[ 

238: \lim_{N \to \infty} \frac{1}{N} \sum_{n=1}^N \delta^{}_{R^n \Z^2}

239: (\varphi) = \sum_{r \in \D^{}_{\smc}}

240: \eta^{}_{\smc} (r) \mu^{}_r (\varphi),

241: \]  

242: where the measure $\mu^{}_r$ is the normalised uniform distribution on

243: $\partial B^{}_r (0) = \{ x \in \R^2 : |x| =r \}$, with

244: $\mu^{}_0=\delta^{}_0$. This establishes the following result.

245: 

246: \begin{prop}

247: The sequence $(\omega^{}_N)^{}_{N \ge 1}$ of \eqref{eq:sumrg}

248: converges in the vague topology, and

249: \[ 

250: \omega:= \lim_{N \to \infty} \omega^{}_N  = \sum_{r \in

251:     \D^{}_{\smc}} 

252:    \eta^{}_{\smc} (r) \, \mu^{}_r, 

253: \]

254: with shelling numbers $\eta^{}_{\smc}(r)$ and

255: probability measures $\mu^{}_r$ as introduced above.\qed

256: \end{prop}

257: 

258: 

259: It is obvious that the limit is, at the same time, also a limit of

260: tempered distributions, i.e., a limit in $\Sc'(\R^2)$. As the Fourier

261: transform is continuous on $\Sc'(\R^2)$, one has

262: \[ 

263: \widehat{\omega} = \Bigl( \lim_{N \to \infty} \omega^{}_N

264: \Bigr)^{\widehat{}}  = \lim_{N \to \infty} \widehat{\omega}^{}_N. 

265: \] 

266: Employing the ordinary PSF \eqref{eq:psf}, one finds

267: \[  

268: \widehat{\omega}^{}_N = \frac{1}{N} \sum_{n=1}^N

269: \widehat{\delta}^{}_{R^n \Z^2} = \frac{1}{N} \sum_{n=1}^N

270: \delta^{}_{(R^n \Z^2)^\ast} = \frac{1}{N} \sum_{n=1}^N \delta^{}_{R^n

271: \Z^2} = \omega^{}_N 

272: \] 

273: where we used that $(R \gG)^{\ast} = R \gG^{\ast}$ for an

274: isometry $R$, and $\dens(R^n \Z^2)=\dens(\Z^2)=1$.  Combining the last

275: two equations, one sees that

276: \[ 

277: \widehat{\omega} = \lim_{N \to \infty} \widehat{\omega}^{}_N = \lim_{N

278:   \to \infty} \omega^{}_N = \omega. 

279: \]

280: 

281: Using polar coordinates, the Fourier transform of $\mu^{}_r$ --- which

282: is an analytic function because $\mu^{}_r$ has compact support --- can

283: be expressed by a Bessel function of the first kind via the following

284: calculation,

285: \begin{equation}\label{eq:foubessel}

286: \begin{split} 

287: \widehat{\mu}^{}_r(k) & = \int_{\R^2} e^{-2\pi i k\cdot x} \dd \mu^{}_r(x) 

288:  = \int_{0}^{\infty} \frac{1}{2\pi} \int_{0}^{2\pi} e^{-2\pi i

289:    |k| \varrho \cos{\varphi}} \dd \varphi \dd \delta^{}_r(\varrho)  \\[2mm]

290: & =  \int_{0}^{\infty} J^{}_0(2\pi|k|\varrho) \dd \delta^{}_r(\varrho) =  

291: J^{}_0 ( 2\pi |k| r),  

292: \end{split}

293: \end{equation}

294: where $J^{}_0(z) = \sum_{\ell=0}^{\infty} \frac{(-1)^\ell}{(\ell!)^2}

295: (\frac{z}{2})^{2\ell}$. This yields the identity

296: \begin{equation} \label{eq:sumbess}

297: \widehat{\omega} = \sum_{ r \in \D^{}_{\smc}}

298: \eta^{}_{\smc}(r) J^{}_0(2 \pi |k|r),  

299: \end{equation}

300: to be understood in the distribution sense. It has the

301: following consequence. 

302: 

303: \begin{cor} \label{corsumbess}

304: With $\eta^{}_{\smc}(r)$ and $\mu^{}_r$ as

305: introduced above, one has

306: \[  

307:     \sum_{r \in \D^{}_{\smc}} \eta^{}_{\smc} (r) \, \mu^{}_r =

308: \Bigl(  \sum_{r \in \D^{}_{\smc}}

309: \eta^{}_{\smc} (r) \, \mu^{}_r \Bigr)^{\widehat{}} =

310: \sum_{r \in \D^{}_{\smc}}

311: \eta^{}_{\smc} (r) J^{}_0(2 \pi |k| r ),

312: \] 

313: where the last expression is to be understood in the distribution

314: sense. \qed

315: \end{cor}

316: 

317: Identities of this type can be viewed as measure-valued

318: generalisations of classic Hardy-Landau-Voronoi summation formulae,

319: see \cite[Sec.\ 4.4]{ik} and references given there for

320: details. However, in view of the rather delicate convergence

321: properties, a direct verification via the PSF \eqref{eq:psf} seems a

322: simpler approach. 

323: 

324: Observe that one can alternatively reduce the problem to one dimension

325: and employ a Hankel transform, compare \cite[Sec.\ 4.4]{ik}. This

326: requires a separate treatment of the transformed measure at the origin

327: in $k$-space. As this looks slightly artificial from the point of view

328: of diffraction, we stick to ordinary Fourier transform.

329: 

330: The version for a general lattice $\gG \subset \R^d$ reads as

331: follows, where, in analogy to $\eta^{}_{\smc}(r)$, 

332: $\eta^{}_{\gG}(r)$ and $\eta^{}_{\gG^{\ast}}(r)$ denote

333: the number of points of $\gG$ and $\gG^{\ast}$ on centred

334: shells $\partial B^{}_r(0)$ of radius $r$. 

335: 

336: \begin{thm}[Radial PSF] \label{radpsf}

337: Let $\gG$ be a lattice of full rank in $\R^d$, with dual lattice

338: $\gG^{\ast}$. If the sets of radii for non-empty shells are

339: $\D^{}_{\gG}$ and $\D^{}_{\gG^{\ast}}$, with shelling

340: numbers $\eta^{}_{\gG}(r)$ and $\eta^{}_{\gG^{\ast}}(r)$,

341: the classical PSF \eqref{eq:psf} has the radial

342: analogue  

343: \begin{equation} \label{eq:radpsf}

344: \Bigl( \sum_{ r \in \D^{}_{\gG}}  \eta^{}_{\gG}(r) \, \mu^{}_r

345: \Bigr)^{\widehat{}} 

346: = \sum_{r \in \D^{}_{\gG} } \eta^{}_{\gG}(r)

347:   \, \widehat{\mu}^{}_r

348: = \dens(\gG) \sum_{r \in  \D^{}_{\gG^{\ast}} }

349: \eta^{}_{\gG^{\ast}}(r) \, \mu^{}_r ,

350: \end{equation}     

351: where $\mu^{}_r$ denotes the uniform probability measure on the sphere of

352: radius $r$ around the origin. 

353: \end{thm}

354: 

355: \begin{proof}

356: Select a sequence of isometries $(R^{}_n)^{}_{n \ge 0}, \, R^{}_n \in \mathrm{SO}(d)$,

357: such that $(R^{}_n x)^{}_{n \ge 0}$, for a fixed $x$ of length 1, is

358: uniformly distributed on the unit sphere ${\mathbb S}^d$. Consider

359: then the sequence of measures defined by 

360: \[ \omega^{}_N = \frac{1}{N} \sum_{n=1}^N \delta^{}_{R^{}_n \gG} . \]

361: The claim now follows from Weyl's lemma and the classical PSF

362: \eqref{eq:psf} in complete analogy to our previous planar example. 

363: \end{proof}

364: 

365: The formula for general $d$ can also be expressed in terms of Bessel

366: functions of the first kind. Here, by standard calculations with

367: spherical coordinates and integral representations of Bessel

368: functions, one obtains

369: \begin{equation} \label{ftmur}

370: \widehat{\mu^{}}_r(k) = \int_{\R^d} e^{-2 \pi i kx} \dd \mu^{}_r(x) =

371: \Gamma\bigl(\frac{d}{2}\bigr)\, \frac{ J^{}_{\frac{d}{2}-1} (2 \pi |k|r)}{(\pi

372:   |k|r)^{\frac{d}{2} -1}},  

373: \end{equation}

374: where $\Gamma(x)$ is the gamma function and $J^{}_{\nu}(z) = (

375: \frac{z}{2})^{\nu} \sum\limits_{\ell=0}^{\infty} 

376: \frac{(-1)^\ell}{\ell! \Gamma(\nu + \ell+1)} (\frac{z}{2})^{2\ell}$.

377: In particular, 

378: \[ 

379: \widehat{\mu}^{}_r(k) = \begin{cases} 

380: \cos(2\pi k r), & \text{if $d=1$}, \\[1mm]

381: J^{}_0(2 \pi |k| r), & \text{if $d=2$}, \\[1mm]

382: \frac{\sin(2 \pi |k| r)}{2 \pi |k| r}, & \text{if $d=3$}. 

383: \end{cases} 

384: \]

385: The analogue of Corollary~\ref{corsumbess} for $d=1$ and 

386: $\gG = \Z$ thus reads 

387: \[ \sum_{m \in \Z} \cos(2 \pi km) = \widehat{\delta}^{}_{\Z} =

388: \delta^{}_{\Z}, \]

389: which is just another form of the ordinary PSF in this case (as radial

390: averaging is trivial here). Figure~\ref{j123} shows \eqref{ftmur} for

391: $r=1$ and various values of $d$. 

392: 

393: \begin{figure} 

394: \epsfig{file=murd12345.eps, height=40mm}

395: \caption{\label{j123}

396: A plot of the radial structure of the function $\widehat{\mu}^{}_1(k)$ 

397: of \eqref{ftmur} for dimensions $d=1$ (light grey), $2$, $3$, $4$, and $5$ (black). }

398: \end{figure}

399:  

400: 

401: \section{A simplistic approach to powder diffraction} 

402: 

403: The intensity distribution in powder diffraction emerges from a

404: collection of grains in random and mutually uncorrelated

405: orientations. Its precise theoretical description is difficult,

406: compare \cite{war} and references therein.

407: 

408: Here, we look into a rather simplistic approach that nevertheless

409: captures the essence of the diffraction image. For simplicity, and for

410: comparison with related pinwheel patterns, we explain this for the

411: square lattice $\Z^2$. Instead of working with grains of finite size,

412: we consider the superposition of entire lattices, with appropriate

413: weights. Moreover, we make the restriction that there is a common

414: rotation centre for all lattices, which we choose to be the origin. As

415: a first step, let us take a look at $\Z^2 \cup R \Z^2$, where $R \in

416: \mathrm{SO}(2)$ is a generic rotation (by which we mean that it is

417: \emph{not} an element of the group of coincidence rotations

418: $\mathrm{SOC}(\Z^2) = \mathrm{SO}(2,\Q)$, see \cite{bg} for more on

419: this concept).

420: 

421: \begin{lem} \label{diff2gitter}

422: Let $R \in \mathrm{SO}(2)$ be a generic rotation, so that $\Z^2 \cap R\Z^2 =

423: \{0\}$. Then, the autocorrelation of\/ 

424: $\omega = \frac{1}{2} (\delta^{}_{\Z^2} + \delta^{}_{R \Z^2})$ is

425: \begin{equation} \label {eq:autocor2}

426: \gamma^{}_{\omega} = \frac{1}{4} \delta^{}_{\Z^2} + \frac{1}{4}

427: \delta^{}_{R\Z^2} + \frac{1}{2} \lambda, 

428: \end{equation}

429: where $\lambda$ is the Lebesgue measure in $\R^2$. 

430: The diffraction measure of $\omega$ is 

431: \begin{equation} \label {eq:autocor3}

432: \widehat{\gamma}^{}_{\omega} = \sum_{x \in \Z^2 \cup R\Z^2} I(x)

433: \delta^{}_x 

434: \end{equation}   

435: with $I(0)=1$ and $I(x)= \frac{1}{4}$ for all \ $0 \ne x \in \Z^2 \cup R

436: \Z^2$. 

437: \end{lem}

438: \begin{proof} 

439: The first two terms in \eqref{eq:autocor2} are the autocorrelations of

440: $\frac{1}{2}\Z^2$ and of $\frac{1}{2}R\Z^2$. The third term originates

441: from the cross-correlation between them, where we used the identity

442: \[ 

443: \lambda = \lim_{r \to \infty} \frac{1}{\pi r^2} \sum_{\substack{

444:     x \in \Z^2 \cap B^{}_r(0) \\ y \in R\Z^2 \cap B^{}_r(0)}}

445: \delta^{}_{x-y}, 

446: \] 

447: with the limit being taken in the vague topology. This identity can be

448: derived as follows.  It is easy to see that each square in the square

449: lattice is hit by $x-y$ exactly once, if $x$ is arbitrary but fixed

450: and $y$ runs through $R\Z^2$. (The exception is $y=0$, but we can

451: neglect it since it plays no role in the limit.)  Thus each square in

452: $\Z^2$ is hit with the same frequency in the limit. Since $R$ is a

453: generic rotation, the sequence $(x-y \bmod (1,1))$ is uniformly

454: distributed in the fundamental domain $[0,1)^2$ of $\Z^2$, if $x$ is

455: arbitrary but fixed and $y$ runs through $R \Z^2$. This is a

456: consequence of the uniform distribution of $(n\alpha \bmod 1)^{}_{n \ge

457: 1}$ in $(0,1)$ if $\alpha$ is irrational. Thus, the points in the sum

458: above are uniformly distributed in $\R^2$ in the limit, and the series

459: converges in the vague topology to the Lebesgue measure.

460: 

461: Now, Equation \eqref{eq:autocor3} follows immediately from

462: $\widehat{\lambda} = \delta^{}_0$ and the PSF \eqref{eq:psf}.  

463: \end{proof}

464: 

465: For completeness, let us briefly comment on the situation that

466: $R\in\mathrm{SOC}(\Z^2)$. In this case, $\gT:=\Z^2\cap R\Z^2$ is a

467: sublattice of $\Z^2$ of finite index, which is $1/\dens(\gT)$.  It is

468: possible to show that $\omega=\frac{1}{2}(\delta^{}_{\Z^2}

469: +\delta^{}_{R\Z^2})$ has diffraction

470: \[

471: \widehat{\gamma}^{}_{\omega} = 

472: \delta^{}_{\gT} + \frac{1}{4}\delta^{}_{\Z^2\setminus\gT} +  

473: \frac{1}{4}\delta^{}_{R\Z^2\setminus\gT}.

474: \]

475: The difference to the diffraction formula in Lemma~\ref{diff2gitter}

476: is concentrated on $\gT\setminus\{0\}$, and hence plays no role in our

477: further discussion when $\dens(\gT)\to 0$, which happens under

478: multiple coincidence intersections \cite{bg}. We may thus disregard

479: coincidence rotations for our present purposes.

480: 

481: \begin{figure} 

482: \epsfig{file=sqexact.eps,width=0.7\textwidth}

483: \caption{\label{sqexact}

484: Radial dependence of ring intensities of square lattice powder diffraction.

485: The central intensity is not shown, see text for details.}

486: \end{figure}

487: 

488: \begin{figure} 

489: \epsfig{file=sq25plot.eps,width=0.7\textwidth}

490: \caption{\label{sqnum}

491: Numerical approximation of Figure~\ref{sqexact}, 

492: based on summing Bessel functions, for

493: radii $r\le 25$. The required radial autocorrelation coefficients

494: $\eta^{}_{\smc}(r)$ are taken from Equation~\eqref{etaq}.

495: }

496: \end{figure}

497: 

498: Let us continue by considering multiple intersections.  If all

499: rotations are generic (in the sense that the lattices $R_{i}\Z^2$ and

500: $R_{j}\Z^2$ are distinct apart from the origin) and satisfy a uniform

501: distribution property, we obtain the following result.

502: 

503: \begin{prop}

504: Let $\gG = \Z^2$, $\gG^{}_j = R^{}_j \Z^2$ with

505: generic $R^{}_j$, i.e., $\gG^{}_j \cap \gG^{}_k = \{ 0 \}$

506: for $j \ne k$, and define $\omega = \frac{1}{N} \sum_{j=1}^N

507: \delta^{}_{\gG^{}_j}$. 

508: Then, one has the identities

509: \[ 

510: \gamma^{}_{\omega} = 

511: \frac{N-1}{N} \lambda \; +

512: \; \sum\limits_{j=1}^N \frac{1}{N^2}

513: \delta^{}_{\gG^{}_j}  \quad \mbox{ and }

514: \quad \widehat{\gamma}^{}_{\omega} = \frac{N-1}{N} \delta^{}_0 +

515: \frac{1}{N} \Bigl( \frac{1}{N} \sum\limits_{j=1}^N

516: \widehat{\delta}^{}_{\gG^{}_j}   \Bigr). \]

517: If $( R^{}_j x )^{}_{j \ge 0}$ is uniformly distributed on $\Sb^1$ for

518: some fixed $x \in \Sb^1$, we obtain 

519: \[ \lim_{N\to \infty} \frac{1}{N} \sum_{j=1}^N

520: \widehat{\delta}^{}_{\gG^{}_j} = 

521: \sum_{r \ge 0} \eta^{}_{\smc}(r) \mu^{}_r,   \] 

522: with $\eta^{}_{\smc}(r)$ as in \eqref{etaq}.

523: \end{prop}

524:  

525: \begin{proof}

526: The first statement --- about $\gamma^{}_{\omega}$ --- follows 

527: as in Lemma~\ref{diff2gitter}.  

528: Observing $(\gG^{}_j)^{\ast} = R_j \gG^{\ast}$, this implies 

529: \[ 

530: \widehat{\gamma}^{}_{\omega} =  \sum_{x \in \bigcup_{j}

531:   R_j \gG^{\ast}} I(x) \delta^{}_x, 

532: \]

533: where $I(0)=1$ and $I(x)=\frac{1}{N^2}$ otherwise. Consequently, one

534: finds 

535: \[ 

536: \widehat{\gamma}^{}_{\omega} = \frac{N-1}{N} \delta^{}_0 +

537: \frac{1}{N} \Bigl( \frac{1}{N} \sum_{j=1}^N

538: \widehat{\delta}^{}_{\gG^{}_j} \Bigr). 

539: \] 

540: By Theorem~\ref{radpsf} in connection with Weyl's lemma and the

541: self-duality of $\Z^2$, the term in the brackets converges to $\sum_{r

542: \ge 0} \eta^{}_{\smc}(r) \mu^{}_r$ as $N \to \infty$.

543: \end{proof}

544: 

545: 

546: This simple argument shows that, after discarding the central

547: intensity and multiplying the remainder by $N$, one is left with a

548: circular diffraction pattern in the spirit of the radial PSF in

549: Theorem~\ref{radpsf}. Disregarding the central intensity, the shelling

550: numbers $\eta^{}_{\smc}(r)$ reflect the total intensity, integrated

551: over the rings of radius $r$, in this idealized approach to powder

552: diffraction. Consequently, in a measurement that displays the

553: intensity along a given direction, the resulting radial dependence is

554: given by $\frac{\eta^{}_{\smc}(r)}{2 \pi r}$, compare

555: Figure~\ref{sqexact}. The numerical approximation in

556: Figure~\ref{sqnum}, included for later comparison, shows a strong

557: oscillatory behaviour, as a result of summing Bessel functions.  Note

558: that this approximation disregards positivity in favour of including

559: circular symmetry. The comparison gives a good impression on the

560: overshooting that originates from this approach.

561: 

562: 

563: \section{Application to tilings with statistical circular symmetry} 

564: 

565: \subsection{The pinwheel tiling}

566: The prototile of the pinwheel tiling is a rectangular triangle with

567: side length $1$, $2$ and $\sqrt{5}$.  The smallest angle in $T$ is

568: $\arctan(\frac{1}{2})$.  Here, we choose the prototile $T$ with

569: vertices $(-\frac{1}{2},-\frac{1}{2})$, $(\frac{1}{2},-\frac{1}{2})$,

570: $(-\frac{1}{2},\frac{3}{2})$, and equip $T$ with a control point at

571: $(0,0)$.  Every tile in a pinwheel tiling is either of the form

572: $R^{}_{\alpha} T + x$, or of the form $R^{}_{\alpha} S T +x$ for some

573: $x \in \R^2$, where $S$ denotes the reflection in the horizontal axis

574: and $R^{}_{\alpha}$ rotation through $\alpha$. The substitution

575: $\sigma$ for the pinwheel tiling is shown in Figure~\ref{fig:subst}

576: (left). By the action of $\sigma$, $T$ is expanded, rotated by

577: $-\arctan(\frac{1}{2})$ and dissected into five triangles that are

578: congruent to $T$. Formally, we define $\sigma(T) := \{ Q_1(T)+x_1,

579: Q_2(T)+x_2, Q_3(T)+x_3, Q_4(T)+x_4, Q_5(T)+x_5\}$, with $Q_i \in

580: \mathrm{O}(2)$ and $x_i \in \R^2$. The appropriate choices of $Q_i$

581: and $x_i$ can be derived from the figure. In particular, $\sigma(T)$

582: contains a triangle \emph{equal} to $T$, thus we may choose $Q_1 =

583: \mathbbm{1}$ and $x_1=0$. The substitution $\sigma$ extends in a

584: natural way to all isometric copies $QT+x$ (where $Q \in

585: \mathrm{O}(2)$ and $x \in \R^2$) by

586: $\sigma(QT+x):=Q\sigma(T)+\sqrt{5}x$. Since one of the triangles in

587: $\sigma(T)$ equals $T$, the tiling $PW$ can be defined as a fixed

588: point of the substitution $\sigma$. This results in $\sigma(PW)=PW=

589: \bigcup_{n \ge 0} \sigma^n(T)$. This convention follows \cite{mps} but

590: deviates from \cite{rad}, because it is advantageous for us to include

591: the rotation in the definition of $\sigma$.

592: 

593: \begin{figure}

594: \epsfig{width=\textwidth,file=subst2.eps}

595: \caption{The substitution for the pinwheel tiling (left) and for the

596: kite domino tiling (right). The dashed lines indicate how the dominos

597: have to be dissected to obtain a pinwheel tiling. They are also needed

598: to turn the global substitution rule for $KD$ into a local one for 

599: its hull.

600: \label{fig:subst}}

601: \end{figure}

602: 

603: \subsection{The kite domino tiling}

604: It is advantageous to consider a second substitution which yields a

605: closely related tiling. Consider the two prototiles $K$ and $D$ (for

606: 'kite' and 'domino'), where $D$ is the rectangle with vertices

607: $(-\frac{1}{2},-\frac{1}{2})$, $(\frac{1}{2},-\frac{1}{2})$,

608: $(\frac{1}{2},\frac{3}{2})$, $(-\frac{1}{2},\frac{3}{2})$, and $K$ is

609: the quadrangle with vertices $(-\frac{1}{2},-\frac{1}{2})$,

610: $(\frac{1}{2},-\frac{1}{2})$, $(\frac{11}{10},\frac{3}{10})$,

611: $(-\frac{1}{2},\frac{3}{2})$. Both tiles consist of two copies of the

612: pinwheel triangle $T$, glued together along their long edges. Every

613: pinwheel tiling gives rise to a kite domino tiling by deleting all

614: edges of length $\sqrt{5}$. The substitution for the kite domino

615: tiling is shown in Figure~\ref{fig:subst} (right). Essentially, it is

616: $\sigma^2$ applied to the new prototiles, where $\sigma$ is the

617: substitution for the pinwheel tiling. If $\varrho$ denotes this new

618: substitution, a fixed point is given by $KD = \bigcup_{n \ge 0}

619: \varrho^n(D)$. This is a \emph{global} substitution that works on the

620: specific $D$ defined above, compare \cite{f} for a discussion of the

621: general substitution concept. The two tilings, $PW$ and $KD$, are

622: mutually locally derivable (MLD) in the sense of

623: \cite{bsj}. Essentially, this means that one is obtained from the

624: other by local replacement rules and vice versa.  These rules are

625: evident from Figure~\ref{fig:subst}. 

626: 

627: As $PW$ and $KD$ are MLD, the global substitution $\varrho$ also

628: induces a \emph{local} substitution in the sense of \cite{bs}, where

629: two (oriented) copies of the domino have to be distinguished. Loosely

630: speaking, this follows from the observation that the local surrounding

631: of any domino in $KD$ or one of the other elements in the hull defined

632: by $KD$ determines the type of the domino, and hence how to apply

633: the substitution to it.

634: 

635: 

636: We equip $K$ with control points at $(0,0),(\frac{2}{5},\frac{1}{5})$,

637: and $D$ with control points at $(0,0),(0,1)$. Then, the set of all

638: control points in $KD$ is equal to the set of control points in

639: $PW$. This specific discrete point set is a Delone set, denoted by

640: $\gL$ in the sequel. It is not hard to see that $\gL$ is MLD with both

641: $PW$ and $KD$, even though one direction of the replacement rule is

642: less obvious than the one linking $PW$ and $KD$.

643: 

644: Observe that the relative and absolute frequencies of triangles $T$ in

645: the pinwheel tiling are both equal to $1$. For the relative

646: frequencies, this is clear since there is only one prototile. The

647: absolute frequency then follows from the fact that this prototile has

648: area $1$. Every triangle of $PW$ carries exactly one control point, so

649: $\dens(\gL)=1,$ too. For the kite domino tiling, standard

650: Perron-Frobenius theory yields the relative frequencies of kite

651: (resp.\ domino) as $\frac{5}{11}$ (resp.\ $\frac{6}{11}$). Thus, the

652: absolute frequency of a kite (resp.\ domino) is $\frac{5}{22}$

653: (resp.\ $\frac{6}{22}$). The substitution matrix can be extracted from

654: Figure~\ref{fig:subst}.

655: 

656: \begin{figure}

657: \epsfig{file=kdpatch.eps, width=150mm}

658: \caption{A patch from a kite domino tiling. The control points of the

659:   white tiles are contained in $\Z^2$. The control points in the other

660:   tiles are contained in rotated copies of $\Z^2$, where the common

661:   rotation centre is indicated in the figure. The possible rotations

662:   are described in Claim~\ref{rotgitter}. \label{fig:kdpatch}}

663: \end{figure}

664: 

665: Discrete structures that are MLD lead to dynamical systems that are

666: topologically conjugate \cite{kel}. Therefore, we formulate the hull on

667: the basis of the Delone sets of the control points. Let $\gL$ be one

668: such set, e.g., the set $\gL$ of control points of $PW$ as defined

669: above. Define the orbit closure

670: \[ 

671: \X(\gL) = \overline{ \R^2 + \gL}^{\mathsf{LRT}}  

672: \]

673: in the local rubber topology (LRT) \cite{bl}, which is compact. This

674: topology is slightly different from the one introduced in

675: \cite{rad}. But in the present case, both topologies yield the same

676: hull and are equivalent on $\X(\gL)$.

677: 

678: The following result is well-known \cite{rad,mps}. 

679: 

680: \begin{thm} \label{autoconst}

681: The hull $\, \X(\gL)$ is \ $\Sb^1$-symmetric. Moreover, $(\X(\gL),

682: \R^2)$ is a strictly ergodic dynamical system, i.e., it is uniquely

683: ergodic and minimal.  All elements of\/ $\X(\gL)$ possess the same

684: autocorrelation measure $\gamma = \gamma^{}_{\gL}$ and the same

685: diffraction measure $\widehat{\gamma}^{}_{\gL}$. Both measures

686: are\/ $\Sb^1$-symmetric. \qed

687: \end{thm}

688: 

689: In this sense, speaking of the diffraction of the pinwheel tiling or

690: its control points has a unique meaning. At this point, we know that

691: \[ 

692: \widehat{\gamma} = \left( \dens(\gL) \right)^2 \delta^{}_0 +

693: (\widehat{\gamma})^{}_{\mathsf{cont}} = \delta^{}_0 +

694: (\widehat{\gamma})^{}_{\mathsf{cont}} 

695: \]

696: because a translation bounded circularly symmetric measure cannot

697: contain Bragg peaks other than the trivial one at $k=0$.  That the

698: intensity coefficient of $\delta^{}_0$ is $\left( \dens(\gL)

699: \right)^2$, hence $1$ in our case, is a consequence of the equation

700: \[ 

701: \widehat{\gamma}(\{0\}) = \lim_{r \to \infty} \frac{1}{\pi r^2}

702: \, \gamma( B^{}_r (0)) = \left( \lim_{r \to \infty} \frac{1}{\pi r^2}

703: \, \delta^{}_{\gL} ( B^{}_r (0)) \right)^2 = \left( \dens( \gL )

704: \right)^2. 

705: \] 

706: It can be proved by means of a Fourier concentration argument in

707: connection with results from \cite{al}; alternatively, see \cite{hof}

708: for another approach and \cite{mps} for details on the case at hand.

709: It remains to determine the structure of

710: $(\widehat{\gamma})^{}_{\mathsf{cont}}$ more closely, in particular

711: the separation into singular and absolutely continuous components.

712: 

713: \subsection{Results and observations}

714: 

715: The following claims and conjectures are stated for the particular

716: tiling $KD$ or for its set of control points $\gL$. Since mutual local

717: derivability extends to entire hulls, we remain in the situation of

718: Theorem~\ref{autoconst}. Proofs of the claims are either outlined here

719: or will appear in \cite{bfg}. Let us start with our main observation.

720: 

721: \begin{conj} \label{mainconj}

722: The diffraction measure $\widehat{\gamma}$ of the pinwheel tiling is of the

723: form  

724: \[ 

725: \widehat{\gamma} = \delta^{}_0 + (\widehat{\gamma})^{}_{\mathsf{sc}} +

726: (\widehat{\gamma})^{}_{\mathsf{ac}}.

727: \]

728: There is a countable set $\D^*\subset\R_{\ge 0}$ such that

729: \[ 

730: (\widehat{\gamma})^{}_{\mathsf{sc}} = \sum_{r \in \D^*\setminus\{0\}} 

731: I(r) \, \mu^{}_r. 

732: \]

733: The set $\D^*$ seems to be locally finite, i.e., discrete and closed.

734: Moreover, $(\widehat{\gamma})^{}_{\mathsf{ac}}$ seems to be 

735: non-vanishing.

736: \end{conj}

737: 

738: The set $\D^*$ is a subset of another set, $\D= \D^{}_{\gL}$, which

739: shows up in the determination of the autocorrelation and is described

740: in more detail below.

741: 

742: \begin{claim} \label{rotgitter}

743: Let $\gL$ be the set of all control points of $KD$. Then, $\gL$ is a

744: Delone subset of a countable union of rotated square lattices. More

745: precisely, with $\theta = 2 \arctan(\frac{1}{2})$,   

746: \[ \gL \subset  \bigcup_{k \in \Z}  R^{}_{k \theta}(\Z^2). \]

747: \end{claim}

748: 

749: Using the geometry of the kite domino tiling, it is not difficult to

750: establish this property. Figure~\ref{fig:kdpatch} may serve as a

751: visualisation on a small scale.  This structure suggests that the

752: diffraction of the pinwheel tiling might share some features with that

753: of the powder diffraction of $\Z^{2}$, see the comment after the

754: proof of Lemma~\ref{diff2gitter}. Since $R^{}_{\theta}$ acts as

755: multiplication by $\frac{1}{5} \bigl( \begin{smallmatrix} 3 & -4 \\ 4

756: & 3 \end{smallmatrix} \bigr)$, the next result is immediate.  Note

757: that $R^{}_{\theta}$ is well known from the coincidence site lattice

758: problem of the square lattice, compare \cite{bg}, and has played a

759: crucial rule in the explicit calculations in \cite{mps}.

760:  

761: \begin{claim} \label{distsupset}

762: Let $\gL$ be the set of all control points of $KD$. It satisfies

763: \[ 

764: \gL \subset \bigl\{ (\textstyle{\frac{n}{5^{k}}, \frac{m}{5^{k}}}) :

765: m,n \in \Z, \; k\in\N \bigr\}, 

766: \]  

767: and the distance set $\D = \D^{}_{\gL} :=\{ |x-y| : x, y \in

768: \gL \}$ is a subset of  $\{ \sqrt{ \frac{p^2+q^2}{5^{\ell}} } :

769: p,q,\ell \in \N \}$. It is the same set for all elements of\/

770: $\X(\gL)$. \qed

771: \end{claim}

772: 

773: We believe that this subset relation is quite sharp, i.e., that the

774: difference between the two sets is not too large, in the following sense. 

775: 

776: \begin{claim} \label{allvalues}

777: All values $r = \sqrt{p^2+q^2}$ with $p,q \in \N$ occur in

778: $\D$. Moreover, for each $\ell \in \N$, there are infinitely many $p,q

779: \in \N$ such that $\sqrt{\frac{p^2+q^2}{5^{\ell}}}$ is contained in

780: $\D$.

781: \end{claim}

782: 

783: In fact, numerical computations suggest the following, stronger

784: property. 

785: 

786: \begin{conj}\label{conj2}

787: For each $\ell \in \N$, $\D$ contains all but finitely many

788: values of the form $\sqrt{\frac{p^2+q^2}{5^{\ell}}}$.

789: \end{conj}

790: 

791: In the sequel, the absolute frequencies of distances are of interest,

792: wherefore we define the \emph{radial autocorrelation function}

793: \begin{equation} \label{eq:etar}

794: \eta(r) = \lim_{R \to \infty} \frac{1}{\vol(B^{}_R)} \sum_{

795:   \substack{ x,y \in \gL\cap B^{}_R (0) \\ |x-y| =r  } } 1. 

796: \end{equation}

797: The limit exists due to unique ergodicity, see Theorem

798: \ref{autoconst}. This permits us to write the autocorrelation of $\gL$

799: --- and hence of $\X(\gL)$ --- as

800: \begin{equation} \label{eq:sumrind} 

801:  \gamma = \sum_{r \in \D} \eta(r) \, \mu^{}_r,   

802: \end{equation} 

803: with $\mu^{}_r$ as in Section~\ref{sec:rpsf}. This follows from

804: Theorem~\ref{autoconst} together with Claim~\ref{allvalues}.  Note

805: that $\gL$ is repetitive \cite{mps} (defined up to congruence), so

806: that all $\eta(r)$ in \eqref{eq:sumrind} are strictly positive, by the

807: definition of $\D$ in Claim~\ref{distsupset}.  Clearly, $\gamma$ is

808: both a positive and positive definite measure, and a tempered

809: distribution on $\R^2$.

810: 

811: The following theorem holds in more generality than the context

812: of this paper. In fact, it holds for all substitution tilings which are

813: of \emph{finite local complexity} (FLC) and \emph{self-similar}. The

814: latter means that $\lambda T^{}_i = \bigcup_{T \in \sigma(T^{}_i)} T$. Roughly

815: speaking, this means that the support of the substitution of each

816: prototile $T^{}_i$ is similar to $T^{}_i$. Finite local complexity means

817: that, for some $R>0$, the tiling contains only finitely many local

818: patches of diameter less than $R$, up to congruence. Note that FLC is

819: frequently defined with respect to translations, whereas we define FLC

820: here with respect to congruence, because of the nature of the pinwheel

821: tiling. Both properties, FLC and self-similarity,

822: hold for the majority of substitution tilings in the literature

823: \cite{fh}.

824: 

825: \begin{thm} \label{qlambdad}

826: In any self-similar substitution tiling of finite local complexity

827: {\rm (}w.r.t.\ congruence{\rm )} with substitution factor $\lambda$,

828: all relative frequencies are contained in $\Q(\lambda^d)$.  Moreover,

829: the tiling can be scaled such that all absolute frequencies are

830: contained in $\Q(\lambda^d)$ as well.  In particular, if $\lambda^d$

831: is an integer, and if the tiling is appropriately scaled, all

832: relative and absolute frequencies are rational.

833: \end{thm}

834: 

835: For a proof, we refer to \cite{bfg}. 

836: This theorem, applied to the pinwheel tiling or the kite domino

837: tiling, yields that all frequencies are rational. 

838: 

839: \begin{claim} All values of $r^2\le 5$ together with 

840:   $\eta(r)$ are given in the following table. Values marked with an

841:   asterisk are numerically based conjectures, all

842:   other values are exact. 

843: \[ 

844: \begin{array}{l||c|c|c|c|c|c|c|c|c|c|c|c|c}

845: \rule[-2mm]{0mm}{7mm} r^2 & 0 & \frac{1}{5} & 1 & \frac{8}{5} &

846: \frac{9}{5} & \frac{49}{25} & 2 & \frac{13}{5} & \frac{81}{25} &

847: \frac{17}{5} & 4 & \frac{113}{25} & 5 \\ \hline 

848: \rule[-2mm]{0mm}{7mm} \eta(r) & 1 & \frac{5}{11} &

849: \frac{439}{165} & \frac{1}{2} & \frac{67}{165} & \frac{4}{165}

850: &  \frac{7}{2}^{\ast} & \frac{142}{165} & \frac{4}{165} &

851: \frac{10}{11}^{\ast} &  3^{\ast} &  \frac{8}{165}^{\ast} &

852: \frac{73}{15}^{\ast} 

853: \end{array}

854: \] 

855: \end{claim}\smallskip

856: 

857: The value $\eta(0)$ is the absolute frequency of the control points.

858: Since each triangle in the pinwheel tiling has area $1$, we have

859: $\eta(0)=1$.  The distance $|x-y|=1/\sqrt{5}$ occurs precisely once

860: within each kite. Kites have absolute frequency $\frac{5}{22}$, but

861: the distance $|x-y|$ has to be counted twice in view of formula

862: \eqref{eq:etar}, hence $\eta(1/\sqrt{5})=\frac{5}{11}$. The other

863: exact values contained in the table can be established similarly, but

864: require more sophistication.  With some further effort, one can

865: determine the \emph{frequency module} of $\gL$, which is the $\Z$-span

866: of the absolute frequencies of all finite subsets of $\gL$, the latter

867: standardised to having density $1$ (which is our natural setting here).

868: 

869: \begin{figure}

870: \epsfig{file=vertstarfreq.eps}

871: \caption{The 11 vertex stars of the pinwheel tiling (up to

872:   congruence) and their absolute frequencies (i.e., frequencies per unit

873:   area). \label{fig:vertstars}} 

874: \end{figure}

875: 

876: \begin{claim}

877: The frequency module of $\gL$, and hence of\/ $\X(\gL)$, is 

878: $\F= \{ \frac{m}{264 \, \cdot \,  5^{\ell}} \, | \, m \in \Z, \; \ell

879: \in \N \}$. In particular, it is countably, though not finitely generated. 

880: \end{claim}

881: 

882: This result is derived from the absolute frequencies of the vertex

883: stars in the pinwheel tiling, see Figure~\ref{fig:vertstars}. These,

884: in turn, can be derived from the frequencies of kites and darts in the

885: kite domino tiling. Details will be given in \cite{bfg}.  

886: 

887: We proceed by considering the Fourier transform in relation to the

888: distance set $\D$. Since the Fourier transform is continuous

889: on the space $\Sc'(\R^2)$  of tempered distributions,

890: \eqref{eq:sumrind} becomes  

891: \begin{equation}\label{eq:pinbessel}

892: \widehat{\gamma}^{}_{\gL} = \Bigl( \sum_{ r \in \D}  \eta(r)

893: \, \mu^{}_r \Bigr)^{\widehat{}} = \sum_{r \in \D }

894: \eta(r) \, \widehat{\mu}^{}_r, 

895: \end{equation} 

896: where the sum is to be understood in the distribution sense. By

897: Bochner's theorem, $\widehat{\gamma}^{}_{\gL}$ is again a positive

898: and positive definite measure, and the equation can be understood as a

899: vague limit as well. 

900: 

901: Let us apply Theorem~\ref{radpsf} to the pinwheel pattern $\gL$.

902: According to Claim~\ref{distsupset}, the distances in $\D$ arise from

903: lattices of the form $\frac{1}{5^{\ell/2}}\Z^2$ with $\ell\in\N$.  The

904: corresponding dual lattices, which are relevant for the diffraction

905: analysis, are $5^{\ell/2}\Z^2$, which have distance sets

906: $5^{\ell/2}\D^{}_{\smc}$. Since $\Z^2$ contains a lattice congruent to

907: $\sqrt{5} \Z^2$, it follows that $5^{\ell / 2} \D^{}_{\smc} \subset

908: \D^{}_{\smc}$ for all $\ell \in \N$, and that $\D^{}_{\smc}$ contains

909: all distances that seem relevant from this point of view. At this

910: stage, we have not found a compelling argument to exclude the

911: possibility that certain subsets of $\gL$ contribute relevant distances

912: in a coherent fashion, and thus --  by unique ergodicity -- further

913: rings $I(r) \mu_r$ (with $r\not\in\D^{}_{\smc}$) to

914: $\widehat{\gamma}$. However, on the basis of

915: Theorem~\ref{radpsf}, it is at least plausible that the set $\D^*$ is

916: closely related with $\D^{}_{\smc}$, or even equal to it. In this

917: case, $\widehat{\gamma}^{}_{\gL}$ would show rings $I(r) \, \mu^{}_r$

918: for $r \in \D^{}_{\smc}$ only. According to Claim \ref{allvalues}, all

919: distances have positive frequencies, so that this assumption would

920: lead to

921: \[ 

922: ( \widehat{\gamma})^{}_{\mathsf{sing}} = \sum_{r \in

923:   \D^{}_{\smc}} I(r) \, \mu^{}_r = \delta^{}_0 +

924: ( \widehat{\gamma})^{}_{\mathsf{sc}} = \delta^{}_0 +  \sum_{0 < r \in

925:   \D^{}_{\smc}} I(r) \, \mu^{}_r 

926: \]

927: with $I(0)=1$ and $I(r)>0$ for all $r \in \D^{}_{\smc}$. 

928: 

929: Let us continue our discussion on the basis of this hypothesis.

930: If Observation~\ref{conj2} holds, then, for each $\ell \ge 0$, only

931: finitely many distances of type $\sqrt{\frac{p^2+q^2}{5^{\ell}}}$ are

932: missing, each one contributing an absolutely continuous part to

933: $\widehat{\gamma}$. Since $\widehat{\gamma}$ exists as a translation

934: bounded measure, and since $\widehat{\gamma}^{}_{\mathsf{sing}}$ is

935: already covered by the results above, every additional contribution

936: has to contribute to $\widehat{\gamma}^{}_{\mathsf{ac}}$. This

937: motivates the conjecture that $(\widehat{\gamma})^{}_{\mathsf{ac}} \ne

938: 0$.

939: 

940: \begin{figure} 

941: \epsfig{file=pin5plot.eps,width=0.7\textwidth}

942: \caption{\label{pin5} Numerical approximation of the radial intensity

943: dependence of the pinwheel diffraction pattern. It is based on Equations

944: \eqref{eq:foubessel} and \eqref{eq:pinbessel}, disregarding contributions

945: to the central intensity. The radial autocorrelation coefficients $\eta(r)$ 

946: are estimated from a patch of radius of about $56$ (thus effectively using the

947: fifth iteration of the substitution $\sigma$). The vertical scale is arbitrary

948: in the sense that it has no meaning without local averaging and integration.}

949: \end{figure}

950: 

951: \begin{figure} 

952: \epsfig{file=zoom6plot.eps,width=0.5\textwidth}

953: \caption{\label{zoom6} Detailed view of the shoulder to the left of

954: the peak at $k=1$, calculated from the sixth iteration of the

955: substitution rule, with the large peak intensity truncated.}

956: \end{figure}

957: 

958: Observation~\ref{mainconj} is also supported by numerical computations

959: of $\widehat{\gamma}$. A naive numerical analysis of a large finite

960: portion of $\gL$ would yield no relevant result due to the nature of

961: the pinwheel pattern. In particular, the number of orientations grows

962: only logarithmically with the radius, while the number of tiles grows

963: quadratically. The results above allow a more meaningful

964: computation. In particular, by employing the $\Sb^{1}$-symmetry, the

965: problem becomes one-dimensional, and by \eqref{eq:sumbess}, the

966: Fourier transform is expressed explicitly as a sum of weighted Bessel

967: functions.  As before, this approach disregards positivity of the

968: intensity function, and the result displayed in Figure~\ref{pin5}

969: shows strong oscillations with significant overshooting, similar in

970: kind to the ones observed in Figure~\ref{sqnum}. No smoothing of any

971: kind was used.

972: 

973: 

974: A comparison of the two diffraction images (Figures \ref{sqnum} versus

975: \ref{pin5}) also supports our claim about the possible radii of

976: pinwheel diffraction rings. One noticeable difference is the higher

977: peak at $k=\sqrt{5}$, which is due to the pairs of points in $\gL$

978: with distance $1/\sqrt{5}$. The existence of positive shoulders in

979: Figure~\ref{pin5} (such as that to the left of the first peak at

980: $k=1$, a blow-up of which is shown in Figure~\ref{zoom6}) is another

981: significant difference to Figure~\ref{sqnum}, and is one of the

982: reasons why we expect a non-vanishing radially continuous

983: contribution, hence giving an absolutely continuous component to

984: $\widehat{\gamma}$.

985: 

986: 

987: 

988: \section*{Acknowledgements}

989: It is a pleasure to thank Friedrich G\"otze, Robert V.\ Moody, Thomas

990: Proffen and Anthony Quas for helpful discussions. This work was

991: supported by the German Research Council (DFG) within the

992: Collaborative Research Center 701, and by EPSRC via Grant EP/D058465.

993: 

994: \begin{thebibliography}{99}

995: 

996: \bibitem{baa}

997: M.\ Baake:\

998: \emph{Mathematical Diffraction Theory in Euclidean Spaces},

999: lecture notes, EPFL, Lausanne (2005).

1000: 

1001: \bibitem{bfg}

1002: M.\ Baake, D. Frettl\"oh, U.\ Grimm:\

1003: in preparation.

1004: 

1005: \bibitem{bg}

1006: M.\ Baake, U.\ Grimm:\

1007: Multiple planar coincidences with $N$-fold symmetry,

1008: \emph{Z.\ Krist.} {\bf 221} (2006) 571--581;\newline 

1009: {\tt math.MG/0511306}.

1010: 

1011: \bibitem{bl}

1012: M.\ Baake, D.\ Lenz:\

1013: Dynamical systems on translation bounded measures:\ pure point

1014: dynamical and diffraction spectra,

1015: \emph{Ergodic Theory Dyn.\ Syst.} {\bf 24} (2004) 1867--1893;

1016: {\tt math.DS/0302061}.

1017: 

1018: \bibitem{bs}

1019: M.\ Baake and M.\ Schlottmann:\

1020: Geometric aspects of tilings and equivalence concepts,

1021: in: Proc.\ of the 5th Intern.\ Conf.\ on Quasicrystals,

1022: eds.\ C.\ Janot and R.\ Mosseri, World Scientific,

1023: Singapore (1995), pp.\ 15--21.

1024: 

1025: \bibitem{bsj}

1026: M.\ Baake, M.\ Schlottmann, P.\thinspace D.\ Jarvis:\ 

1027: Quasiperiodic tilings with tenfold symmetry and equivalence with

1028: respect to local derivability, 

1029: \emph{J.\ Phys.\ A:\ Math.\ Gen.} {\bf 24} (1991) 4637--4654.

1030: 

1031: \bibitem{cor1}

1032: A.\ C\'ordoba:\

1033: La formule sommatoire de Poisson,

1034: \emph{C.\ R.\ Acad.\ Sci.\ Paris, S\'er.\ I:\ Math.} {\bf 306} (1988)

1035: 373--376. 

1036: 

1037: \bibitem{cor2}

1038: A.\ C\'ordoba:\

1039: Dirac combs,

1040: \emph{Lett.\ Math.\ Phys.} {\bf 17} (1989) 191--196.

1041: 

1042: \bibitem{f}

1043: D.\ Frettl\"oh:\

1044: Duality of model sets generated by substitutions, 

1045: \emph{Rev.\ Roumaine Math.\ Pures Appl.} {\bf 50} (2005) 619--639;

1046: {\tt math.MG/0601064}.

1047: 

1048: \bibitem{fh}

1049: D.\ Frettl\"oh and E.\ Harriss,

1050: \emph{Tilings Encyclopedia}, available online at:\newline

1051: {\tt http://tilings.math.uni-bielefeld.de/}

1052: 

1053: \bibitem{haw}

1054: G.\thinspace H.\ Hardy, E.\thinspace M.\ Wright:\

1055: \emph{An Introduction to the Theory of Numbers}, 5th ed.,

1056: Clarendon Press, Oxford (1979).

1057: 

1058: \bibitem{hof} 

1059: A.\ Hof:\

1060: On diffraction by aperiodic structures,

1061: \emph{Commun.\ Math.\ Phys.} {\bf 169} (1995) 25--43.  

1062: 

1063: \bibitem{ik}

1064: H.\ Iwaniec, E.\ Kowalski:\

1065: \emph{Analytic Number Theory},

1066: AMS, Providence, RI (2004). 

1067: 

1068: \bibitem{kel}

1069: J.\ Kellendonk:\

1070: Topological equivalence of tilings,

1071: \emph{J.\ Math.\ Phys.} {\bf 38} (1997) 1823--1842;\newline

1072: {\tt cond-mat/9609254}.

1073: 

1074: \bibitem{kn}

1075: L.\ Kuipers, H.\ Niederreiter:\

1076: \emph{Uniform Distribution of Sequences},

1077: Wiley, New York (1974).

1078: 

1079: \bibitem{al}

1080: J.\ Gil de Lamadrid, L.\thinspace N.\ Argabright:\

1081: \emph{Almost Periodic Measures},

1082: Memoirs AMS, vol.\ 85, no.\ 428, Providence, RI (1990). 

1083: 

1084: \bibitem{moo}

1085: R.\thinspace V.\ Moody:\ 

1086: Model sets:\ A survey, in:\

1087: \emph{From Quasicrystals to More Complex Systems},

1088: eds.\ F.~Axel, F.~D\'enoyer, J.\thinspace P.~Gazeau,

1089: EDP Sciences, Les Ulis, and Springer, Berlin (2000), pp.\ 145--166;\newline

1090: {\tt math.MG/0002020}.

1091: 

1092: \bibitem{mps}

1093: R.\thinspace V.\ Moody, D.\ Postnikoff and N.\ Strungaru:\ 

1094: Circular symmetry of pinwheel diffraction,

1095: \emph{Ann.\ Henri Poincar\'e} {\bf 7} (2006) 711--730.

1096: 

1097: \bibitem{rad}

1098: C.\ Radin:\

1099: Aperiodic tilings, ergodic theory and rotations, in:\

1100: \emph{The Mathematics of Long-Range Aperiodic Order}, 

1101: ed.\ R.\thinspace V.\ Moody,

1102: NATO ASI C 489, Kluwer, Dordrecht (1997), pp.\ 499--519.

1103: 

1104: \bibitem{sch}

1105: L.~Schwartz:\

1106: \emph{Th\'eorie des Distributions}, 

1107: rev.\ ed., Hermann, Paris (1998).

1108: 

1109: \bibitem{war}

1110: B.\thinspace E.\ Warren:\

1111: \emph{X-ray Diffraction}, reprint, Dover, New York (1990).

1112: 

1113: \end{thebibliography}

1114: \bigskip

1115: 

1116: \parindent 0pt

1117: 

1118: \end{document}

1119: 

1120: 

1121: 

1122: 

1123: