1: \documentclass[twoside]{article} \textheight 9in\textwidth 6.5in
2: \topmargin-3pc\oddsidemargin0in\evensidemargin0in\pagestyle{myheadings}
3: \markboth {\hfill Leonid A. Levin \hfill} { \hfill Aperiodic Tilings \hfill}
4: \usepackage{amssymb,epsfig} \begin{document}
5: \newtheorem{dfn} {Definition} \newtheorem{rem} {Remark}
6: \newtheorem{theorem}{Theorem} \newtheorem{lemma}{Lemma}
7: \newtheorem{cor}{Corollary} \newcommand{\ie}{{\em i.e. }}
8: \newcommand\eg{{\em e.g., }} \newcommand\qed{\rule{1em}{1.5ex}}
9: \newcommand\fd[1]{{\em\bf #1}} \newcommand\df[1]{{\bf\em #1}}
10:
11: \title {\vspace*{-3pc}Aperiodic Tilings: Breaking Translational Symmetry}
12:
13: \author {Leonid A.~Levin\thanks {Supported by NSF grant CCR-0311411.}\\
14: Boston University\thanks {Computer Science department, 111 Cummington
15: St., Boston, MA 02215; (email: Lnd at bu.edu)}} \date{}\maketitle
16:
17: \begin{abstract} Classical results on aperiodic tilings are rather
18: complicated and not widely understood. Below, an alternative approach is
19: discussed in hope to provide additional intuition not apparent in
20: classical works. \end{abstract}
21:
22: \section {Palettes and Tilings}
23:
24: Physical computing media are asymmetric. Their symmetry is broken
25: by irregularities, physical boundaries, external connections, and so on.
26: Such peculiarities, however, are highly variable and extraneous
27: to the fundamental nature of the media. Thus, they are pruned
28: from theoretical models, such as cellular automata, and
29: reliance on them is frowned upon in programming practice.
30:
31: Yet, computation, like many other highly organized activities, is
32: incompatible with perfect symmetry. Some standard mechanisms must assure
33: breaking the symmetry inherent in idealized computing models.
34: A famous example of such mechanisms is aperiodic tiling: hierarchical
35: self-similar constructions, first used for computational purposes in a
36: classical -- although rather complicated -- work \cite{berger} and
37: further developed in \cite{robinson, myers,gurkor72}. \cite{ad-bgg} give
38: a helpful exposition.
39:
40: \begin{dfn} Let $G$ be the grid of unit length edges between integer
41: points on an infinite plane.\par
42: A \fd {tiling} $T$ is its mapping into a finite set of \fd {colors}.
43: Its \fd {crosses} and \fd {tiles} are ordered color combinations of four
44: edges sharing a corner or forming a square, respectively. A \fd{palette}
45: $P$ of $T$ is a set including all its tiles (\fd{+palette} for crosses).
46: We say $P$ with a mapping $f$ of its colors into a smaller color
47: alphabet \fd {enforces} a set $S$ of tilings if replacing colors
48: according to $f$ turns each $P$-tiling into a tiling in $S$.\end{dfn}
49:
50: Turning each edge orthogonally around its center turns $G$ into
51: its dual graph and palettes into +palettes and vice versa.
52: Thus, one can use either type as convenient.
53:
54: \section {2-Adic Coordinates}
55:
56: The set of all tilings with a given palette $P$ has translational
57: symmetry, \ie any shift produces another $P$-tiling. We want a palette
58: that forces a complete spontaneous breaking of this symmetry, \ie
59: prevents individual tilings from being {\em periodic}.
60: Accordingly, each location in a given tiling will be uniquely characterized
61: by a sort of {\em coordinates}. Their infinitely many values cannot be
62: reflected in the finite variety of the tile's colors. They will be
63: represented {\em distributively}, \ie in the colors of the surrounding
64: tiles, and computable from them to any given number of digits.
65:
66: Let us first so distribute the horizontal Cartesian integer coordinates
67: $x=(2i\!+\!1)2^k$ of vertical edges by reflecting one bit $(i\bmod 2)$
68: in their color. We view this bit as the direction of a \df {bracket}. In
69: this 1-dimensional tiling $C_1$, the brackets of the same \df {rank} $k$
70: are equidistant (Figure~\ref{C1-by-rank}).
71:
72: \newcommand\ranks{\centering \fbox{\begin{minipage}{3in}{\begin{tabbing}
73: [ \=\ \ ] \=\ \ [ \=\ \ ] \=\ \ [ \=\ \ ] \=\ \ [ \=\ \
74: ] \=\ \ [ \=\ \ ] \=\ \ [ \=\ \ ] \=\ \ [ \=\ \ ]\+\\
75: \{\>\>\}\>\>\{\>\>\}\>\>\{\>\>\}\>\>\{\+\\\+[\>\>\>\>]\>\>\>\>[\+
76: \\\{\>\>\>\>\>\>\>\>\}\+\+\+\\\>[ \end{tabbing}}\end{minipage}}
77: \caption {Brackets of $C_1$ split by rank. \label {C1-by-rank}}}
78:
79: It is convenient to visualize the bits of even rank, picturing
80: them \}/\{ or red, or dotted, separately from odd, depicted ]/[.
81: The bits of either shape at each side of the origin form a progression
82: of balanced parenthetical expressions, called \df {domains}.
83: Each domain has four \df {grandchildren} of the second lower rank:
84: two within its outermost brackets and one to each side.
85: The two \df {children} have the other shape and are centered at each
86: border of the \df {parent}, thus connecting it with its grandchildren.
87:
88: Handling the vertical coordinate similarly yields a neat 2-dimensional
89: tiling $C$ called \df {central}. Figure~\ref{Boldness-in-C} marks the
90: borders of intersections of its vertical and horizontal domains of equal
91: ranks with a \df {boldness} bit.
92:
93: \begin{figure}[ht]
94: \hspace*{\fill}\parbox[c]{3in}{\ranks
95: \caption{
96: (Right)\hfill
97: Boldness bit: bold lines in $C$.
98: {Courtesy of A. Shen and B. Durand.} \label {Boldness-in-C}}
99: } \hfill\parbox[c]{3in}
100: {\includegraphics[scale=0.6]{ce-bold.eps}}
101: \end{figure}
102:
103: $C_1$ has a special, \ie unmatched, bracket in the origin, directed
104: arbitrarily and unranked. No palette can enforce a set of tilings with
105: unique special points (designated by a Borel function commuting with
106: shifts) since the set of all tilings is compact\footnote
107: {and so has finite shift-invariant measures, \eg
108: defined by condensation points of frequencies of
109: finite configurations in some quasiperiodic tiling}
110: whereas the set of locations of their special point and the group
111: $\mathbb{Z}$ of their shifts is not. We will extend $\mathbb{Z}$ to a
112: compact group and also \df {define ranks} in other tilings, \eg shifted
113: $C_1$, using the following property.
114:
115: \begin{rem}\label{R} A shift by $(2i\!+\!1)2^k$ in $C_1$
116: reverses all brackets of rank $k\!-\!1$, none of lower ranks,
117: and every second bracket of any rank $>k$.\end{rem}
118:
119: Therefore, the shifts by $(2i\!+\!1)2^k$ change our bits only at a $2/2^k$
120: fraction of locations. This fraction can be used as a metric on the
121: group of shifts which can then be completed for it. The result is a
122: remarkable compact group $g$ of 2-adic integers, or \df {2-adics}, acting
123: on a similarly completed set of 2-adic coordinates. A 2-adic $a$ is a
124: formal infinite sum $\sum_{i\ge0} 2^i a_i=\ldots+4a_2+2a_1+a_0$, where
125: $a_i\in\{0,1\}$, viewed as an infinite to the left sequence of bits.
126: The usual algorithms for addition and multiplication make $g$ a ring
127: with $\mathbb{Z}$ as a subring (\eg $-1=\ldots+8+4+2+1$).\footnote
128: {\samepage Odd 2-adics have inverses. This allows extending
129: $g$ to a famous locally compact field with fractions $a/2^i$.}
130:
131: The natural action of $\mathbb{Z}$ (by shifts) can be extended
132: to the action of the whole $g$ on $C_1$ and its images.
133: Indeed, the brackets of rank $k$ are unaffected by terms $a_i$ with
134: $i>k\!+\!1$. Thus, a 2-adic shift $a$ of $C_1$ can be defined as the
135: pointwise limit of the sequence of shifts by integers approximating $a$.
136:
137: With inverse shifts, this sequence diverges for the unranked bracket in
138: the origin of $C_1$ and of its integer shifts.
139: The direction of this bracket is determined not by its location, but by
140: an arbitrary \df {default} included as an additional (external, unmoved
141: by shifts) point in the tilings. The \df {reflection} reverses the
142: default, all brackets, and the signs of their locations.
143: With added reflection, the action of $g$ is free and transitive:
144: each of these tilings can be obtained from any other, \eg from $C_1$.
145: In 2 dimensions we can also add the diagonal reflection
146: exchanging the vertical and horizontal coordinates.\footnote
147: {We allow fewer tilings than \cite{ad-bgg} which
148: permits different shifts at each side of the origin.}
149:
150: \section {Enforcing the Coordinate System with a Palette}
151:
152: \begin{theorem}\label{E} The set of 2-adic shifts and reflections
153: of tiling $C$ can be enforced by a palette. \end{theorem}
154:
155: To prove it, we use multicomponent colors in $T$ which $f$ projects
156: onto their first component -- bracket bit. The second component includes
157: two \df {enforcement} bits that extend $C$ to the enhanced tiling $CE$,
158: with a +palette $ce$ of 7 crosses modulo 8 reflections.\footnote
159: {\cite{robinson} uses only six tiles (with reflections)
160: but colors their corners, in addition to sides.}
161: One of its bits is the already described boldness bit (Figure~\ref
162: {Boldness-in-C}). The other is a \df {pointer} in the direction of the
163: nearest orthogonal line of the same rank. On the (unranked) axes these
164: bits are set by a default central cross.
165:
166: \begin{dfn} A \fd {box} is an \fd {open} $ce$-tiled rectangle,
167: \ie one with the border edges removed.
168: Its \fd {$k$-block} is a square with monochromatic sides that is\footnote
169: {The rest of the requirement is redundant but useful in the proof.}
170: a tile (for $k\!=\!0$) or a combination of four $(k\!-\!1)$-blocks
171: sharing a corner. The four segments connecting the block's center to its
172: sides are called \fd {$(k\!\!-\!\!1) $-medians}.
173: The rest of the open block is called a \fd {frame}.
174: We call a box \fd {$k$-tiled} if removing an outer layer which is
175: thinner than a $k$-block turns it into a box composed of (open at the
176: box border) $k$-blocks.\footnote {$ce$ prevents crossing of monochromatic
177: segments, making decompositions of boxes into blocks unique.}\end{dfn}
178:
179: \begin{lemma}\label{blocks}
180: (i) Borders between open blocks in a box are monochromatic.
181: (ii) All $k$-frame patterns in a box are enclosed in its open $k$-blocks.
182: (iii) All open $k$-blocks are congruent and have equal frames.\end{lemma}
183:
184: \paragraph {Proof.} (i) comes from all $ce$-crosses having one or four
185: inward pointers. (ii,iii) for $k\!>\!2$ follow from $k\!-\!1$ by viewing
186: 1-blocks as tiles. Let 1-blocks $a$ and $b$ be adjacent in a 2-block
187: $c$; $(L,l)$ and $(R,r)$ be pairs of medians of $a$ and $b$ with $l,r$
188: directed to the side $s$ of $c$, $L\!-\!R$ crossing a median $m$ of $c$
189: at a cross $x$. $x$ forces $L,R$ to be both pale or both bold. This
190: forces opposite brackets on $l,r$ which, too, must be both pale or both
191: bold depending on the bracket of $L\!-\!R$. $l,r$ cannot be both bold
192: which would require the pointer of $s$ to agree with their opposite
193: brackets. Thus, all external medians of 2-frames are pale, internal
194: medians bold, their brackets face the frame's center forcing inward
195: pointers on the 1-medians, like $m$. \qed
196:
197: \begin{lemma}\label{I} Any 1-tiled box is $k$-tiled.
198: (Follows from $k\!\!=\!\!2$ case by seeing $(k\!\!-\!\!2)$-blocks as tiles.)
199: \end{lemma}
200:
201: \paragraph {Proof.} The eight colors of edges fix their location
202: in 2-frames, forcing open 1-blocks to alternate in the pattern of
203: 2-frames which, by Lemma~\ref{blocks}, extend to 2-blocks. \qed
204:
205: \begin{cor} Any $2^k\times2^k$ box, extendible to a 3 times wider
206: cocentric 1-tiled box, extends to a $(k\!+\!4)$-block.\end{cor}
207:
208: \paragraph {Induction Basis.} For the simplest enforcement of tiling
209: decomposition into 1-blocks we can use a 2-periodic \df {parity} bit to
210: mark \df {odd} lines carrying 0-medians.
211: All pointers on odd lines point to odd crossing lines, thus forcing a
212: period 2 on them.
213: One needs only to assure an odd line exists. This can be easily done
214: with a \df {parity pointer} on even lines, pointing to a crossing odd
215: line.
216:
217: \paragraph {\hspace*{1pc} Proof of Corollary.} The box is $k$-tiled
218: covering the inner box with four blocks sharing a cross. The extension
219: comes by viewing them as tiles with parity bit reflecting the blocks'
220: orientation and noting that each cross of $C$ with parity appears in its
221: open 4-blocks. \qed
222:
223: {\bf Proof of Theorem~\ref{E}.} Let $T$ be a $ce$-tiling decomposed,
224: for each $k$, into $k$-blocks
225: with equal frames. Then a shift of $CE$ matches $T$ on all lines of
226: rank $<k$.
227: The shifted $CE$ converge pointwise to $T$, except possibly on their
228: (unranked) axes.
229: By Remark~\ref{R}, the shift increments grow in rank, and so sum to one
230: 2-adic shift.
231: Finally, reflections match the brackets on axes. \qed
232:
233: \subsection {Parsimonious Enforcement of the Grid of 1-Blocks}
234:
235: First, we reduce the needed parity colors.
236: A parity pointer on a single edge suffices,
237: so it needs to accompany only one color if we show that $ce$-tilings
238: cannot skip colors.
239: Indeed, all $ce$-crosses are either \df {bends}, \ie have 4 inward
240: pointers, or \df {passes}, \ie have 1.
241: Thus, a third of crosses are bends, up to $O(n)$ accuracy for
242: $n\times n$ boxes.
243: Moreover, all orientations of bends are equally frequent, alternating
244: on each line. \qed
245:
246: Tedious case investigation of~\cite {levitsky} shows $ce$ bits
247: themselves forcing 1-blocks, rendering parity bits redundant.
248: A \df {$k$-bar} is a maximal bold or pale segment, $k$ being its length.
249: $k>1$ and no bold 3-bar exists since it is easy to see that its middle
250: link would be connected by a tile to a 1-bar. Levitsky first proves that
251: each $ce$-tiling has bold 2-bars. Here is a simpler argument for this.
252:
253: A third of crosses are bends, so the average bar length is 3.
254: Absent bold 2-bars, this average would allow positive density only of
255: bold 4-bars and pale 2-bars. Tilings with such bars have period 6 and
256: map onto a $6\times 6$ torus with 3 bold $4\times 4$ squares. But $Z_6$
257: cannot have three disjoint pairs of points of equal parity! \qed
258:
259: The rest of \cite {levitsky} analysis assures bold 2-bars two tiles
260: away at each side of any bold 2-bar. This involves a case-by-case
261: demonstration that no violation can be centered in a $10\times 10$ box.
262: The analysis is laborious but may be verifiable by a computer check.
263:
264: \section {Acknowledgments}
265:
266: These remarks were developed in my attempts to understand the classical
267: constructions of aperiodic tilings while working on~\cite{dls}.
268: My main source of information was \cite{ad-bgg} and its explanations
269: by B. Durand and A. Shen to whom I owe all my knowledge on this topic.
270:
271: \begin {thebibliography} {99}
272:
273: \bibitem[Berger 66]{berger} Robert Berger. The Undecidability
274: of the Domino Problem. {\em Mem.Am.Math.Soc.}, {\bf 66}, 1-72, 1966.
275:
276: \bibitem[Robinson 71]{robinson} Raphael M.~Robinson.
277: Undecidability and Nonperiodicity of Tilings of the Plane.\\
278: {\em Inventiones Mathematicae}, {\bf 12}:177--209, 1971.
279:
280: \bibitem[Myers 74]{myers} Dale Myers. Nonrecursive Tilings of the Plane.
281: ii. {\em J. Symbolic Logic}, {\bf 39(2)}:286--294, 1974.
282:
283: \bibitem[Gurevich Koriakov 72]{gurkor72} Yu.~Gurevich, I.~Koriakov.
284: A Remark on Berger's Paper on the Domino Problem.
285: {\em Siberian Math. J.}, {\bf 13}:459--463, 1972.
286:
287: \bibitem[Allauzen Durand 96]{ad-bgg} Cyril Allauzen, Bruno Durand.
288: Appendix~A: Tiling Problems. pp.~407--420.\\ In Egon Boerger, Erich Graedel,
289: Yuri Gurevich. {\em The Classical Decision Problem.} Springer-Verlag, 1996.
290:
291: \bibitem[Levitsky 04]{levitsky} Alexander Levitsky.
292: {\em Aperiodicheskie Zamoshcheniya s Malym Alfavitom\\ (= Aperiodic Tilings
293: with Small Alphabet)}. In Russian. Master Thesis. Moscow University, 2004.
294:
295: \bibitem[Durand, Levin, Shen 01]{dls}
296: Bruno Durand, Leonid A. Levin, Alexander Shen. Complex Tilings.
297: {\em Proc. 33rd Annual ACM Symp. on Theory of Computing},
298: July 6-8, pp. 732-739. Heraklion, Crete, Greece, 2001.
299:
300: \bibitem[Gacs 01]{gacs} Peter Gacs.
301: Reliable Cellular Automata with Self-Organization.
302: {\em J.Stat.Phys.},\\ {\bf 103(1,2)}:45--267, 2001.
303: Also available at: http://www.arxiv.org/abs/math.PR/0003117 .
304:
305: \end{thebibliography} \end{document}
306: