Limit-(quasi-)periodic point sets as quasicrystals

with p-adic internal spaces

arXiv:math-ph/9901008v1 17 Jan 1999

Michael Baake‡1 , Robert V. Moody2

and Martin Schlottmann2
1
Institut für Theoretische Physik, Universität Tübingen,
Auf der Morgenstelle 14, D-72076 Tübingen, Germany
2
Department of Mathematical Sciences, University of Alberta,
Edmonton, Alberta T6G 2G1, Canada

Dedicated to Peter Kramer on the occasion of his 65th birthday

Abstract. Model sets (or cut and project sets) provide a familiar and commonly
used method of constructing and studying nonperiodic point sets. Here we extend this
method to situations where the internal spaces are no longer Euclidean, but instead
spaces with p-adic topologies or even with mixed Euclidean/p-adic topologies. We
show that a number of well known tilings precisely fit this form, including the chair
tiling and the Robinson square tilings. Thus the scope of the cut and project formalism
is considerably larger than is usually supposed. Applying the powerful consequences
of model sets we derive the diffractive nature of these tilings.

‡ Heisenberg Fellow
p-adic quasicrystals 2

1. Introduction

The cut and project method of constructing nonperiodic point sets, as developed by
Peter Kramer and others in the early eigthies [10, 11, 9, 3, 12], is one of the basic tools
in the mathematical study of quasicrystals and aperiodic order. The intuition behind
their use is that quasiperiodic point sets, such as those arising in many nonperiodic
tilings and also in the diffraction patterns of physical quasicrystals, may be viewed
as arising from the projection of lattices in some higher dimensional spaces. Thus the
physical space is complemented by an internal space (possibly of some other dimension),
a lattice is located in the combined physical-internal space pair, and the projection maps
are used to create a cut and project scheme.
The same type of mathematical structure had also arisen (before the recent
excitement about quasicrystals, and in a very different context) in the work of Yves
Meyer [13] in which the formalism is expressed entirely in terms of locally compact
Abelian groups. In [14, 19] these ideas were taken up and extended in the context of
aperiodic order, with the result that a considerable amount of the mathematical theory
underlying these cut and project sets (or model sets) can now be seen to hold in great
generality. Up to now, however, no attempt has actually been made to see to what extent
existing aperiodic structures might be explained in terms of these more general types of
spaces. In this paper we address this question, showing that a number of familiar tilings
and substitution systems, so far not contained under the aegis of the cut and project
formalism, are in fact based on internal spaces with non-Euclidean topologies, namely
p-adic topologies or mixed Euclidean/p-adic topologies.
There is a variety of discrete structures known that display a pure point diffraction
spectrum, i.e., the Fourier transform of their autocorrelation (which is a positive
measure) is pure point, compare [8, 22]. Among the known examples are model sets, but
also certain inflation-generated point sets and tilings, e.g. the chair or the sphinx tiling
[22]. They are limit-periodic structures with a countably, but not finitely, generated
Fourier module – so, they cannot be described in the “conventional” cut and project
setup where the internal space is Euclidean.
It is the aim of this contribution to start to develop a proper generalization of
the projection method, in the spirit of Meyer, to include such limit-periodic and even
limit-quasiperiodic sets. Here, we explain, in an illustrative fashion, how this works. A
detailed approach to model sets over arbitrary internal groups, and especially to aspects
of diffraction, will appear in [20].
Let us recall the notion of a cut and project scheme. By definition, this consists of
p-adic quasicrystals 3

a collection of spaces and mappings:

π π
Rd ←−
1
Rd × G −→
2
G
∪ (1)
L

where Rd is a real Euclidean space and G is some locally compact Abelian group, π1
and π2 are the projection maps onto them, and L ⊂ Rd × G is a lattice, i.e., a discrete
subgroup such that the quotient group (Rd × G)/L is compact. We assume that π1 |L is
injective and that π2 (L) is dense in G. We call Rd (resp. G) the physical (resp. internal)
space.
In this definition, we have already oriented the situation to physical applications by
assuming that the physical space is indeed a real Euclidean space. On the other hand,
allowing G to be an arbitrary locally compact Abelian group is precisely the point at
which we are going beyond the usual situation of an internal space that is also Euclidean.
Given any subset Ω ⊂ G, we define a corresponding set Λ(Ω) ⊂ Rd by
Λ(Ω) = {π1 (x) | x ∈ L, π2 (x) ∈ Ω} . (2)
We call such a set Λ a model set (or cut and project set) if the following condition is
fulfilled,
W1 Ω = int(Ω) 6= ∅ is compact.
Furthermore, we are mainly interested in the situation that the boundary of Ω does
not contain any points of π2 (L). If this is the case, we call Λ(Ω) regular. The importance
of regular model sets is that they are necessarily repetitive [19]. Note that, if Ω fulfils
W1, ∂Ω (which equals Ω\int(Ω)) is nowhere dense and hence a meager set. Then, it
follows from the Baire category theorem that no countable union of translates of ∂Ω
can cover G which is a Baire space. In particular, it is always possible to choose a shift
c ∈ G such that the boundary of c + Ω satisfies the additional regularity condition.
In the sequel, showing that a certain set is a model set actually means, more
precisely, to show that there exist G, L and Ω subject to the above conditions such
that Λ(Ω) is regular and locally isomorphic to the given set (for terminology, see [1, 19]
and references therein). This way, our results are valid for entire LI-classes, even if, for
simplicity, we only talk of single tilings.

2. p-adic topologies and inverse limits of finite groups

Let p be a prime number in the integers Z. Using p we can define a metric on the rational
numbers Q, and by restriction on Z, in the following way. For each a ∈ Z, we define its
p-value, νp (a), as the largest exponent k for which pk divides a (with νp (0) := ∞). This
p-adic quasicrystals 4

function is extended to the p-adic valuation νp : Q −→ Z by νp (a/b) := νp (a) − νp (b) for

all rational numbers a/b. We now define the “distance” between two rational numbers
x, y as d(x, y) = p−νp (y−x) .
It is not hard to see that this does define a metric on Q, in which closeness to 0
is equivalent to high divisibility by the prime p. The completion of the rationals under
this topology is the field of p-adic numbers Q cp and the completion of Z is the subring of
p-adic integers, Zcp . Any p-adic integer has a unique expansion (as a convergent series)
P∞
in the form n=0 an pn where the ap are integers in the range 0 ≤ ap < p. The topologies
defined by such metrics have rather counter-intuitive properties. For example, for each
cp , the set of elements of Z
non-negative integer k, the set pk · Z cp divisible by pk , is the ball
of radius p−k and is clopen, i.e. both open and closed. Z cp , seen as a topological space,
is both compact and totally disconnected. In particular, Q cp and Z cp are locally compact
Abelian groups under addition. Thus, we can use Z cp to construct interesting cut and
project schemes for Rd simply by taking G := Z cp and L = Zd embedded diagonally into
Rd × Zcp . For more on p-adic numbers and other totally disconnected groups, the reader
may consult [15, 21, 7, 2].
There is another description of Z cp which is more revealing of its appearance in the
context of self-similarity and generalizes what we have just done. Let
F1 ← F2 ← F3 ← . . . (3)
be an inverse system of finite Abelian groups, i.e. each Fi is a finite Abelian group (with
discrete toplogy) and the arrows represent surjective group homomorphisms. Define the
set of compatible sequences
←−
F := {x̃ = (x1 , x2 , . . .) | xi ∈ Fi , xi ←| xi+1 , i ∈ N} . (4)
←−
F is given the structure of a group by component-wise addition. It is structured as a
Q
topological group by the induced topology from the product ∞ i=1 Fi . Equivalently, the
subgroups
←− ←
−
Fn = {x̃ ∈ F | x1 = x2 = . . . = xn−1 = 0} (5)
←−
form a subbase of open neighbourhoods of 0 in F . Since [F : Fn ] is finite, the subgroups
←− ←−
F n are also closed. With this topology, F is a compact totally disconnected Abelian
group (so, in particular, a locally compact Abelian group). Groups of this type are
called profinite groups.
As an example, for each prime number p, we can construct Z cp by the inverse system
←−
Z p : Z/pZ ← Z/p2 Z ← Z/p3 Z ← . . . (6)
The relevance to the work here is this: if θ : L → L is an injective homomorphism
that is a self-similarity of Λ, then there is a clear distinction between the case that
p-adic quasicrystals 5

θ(L) = L (θ is a “unit”) and the case that θ(L) ⊂ L, but θ(L) 6= L. In the latter case,
[L : θ(L)] is finite and we have the inverse system
←−
L (θ) : L/θ(L) ← L/θ2 (L) ← L/θ3 (L) ← . . . (7)
←
−
The compact group L is invariant under the action of θ. Note that it contains a
canonical copy of L itself via the mapping
x 7→ ((x mod θ(L)) ←| (x mod θ2 (L)) ←| (x mod θ3 (L)) ←| . . .) . (8)
← −
Again, we obtain a cut and project scheme via the diagonal embedding of L in Rm × L .
Let us now turn to some applications.

3. A limit-periodic substitution system

Consider the primitive three-letter substitution system

a → ab b → abc c → abcc (9)
The standard analysis of the corresponding substitution matrix [17] shows that a proper
geometric realization demands length ratios ℓ(a) : ℓ(b) : ℓ(c) = 1 : 2 : 3, while all three
letters finally occur with equal frequency 1/3.
To obtain a bidirectional infinite sequence that is a fixed point, we may start with
the pair c|a and keep on applying the substitution rule:
. . . ababcabccabcc|ababcababcabccababcababcabcc . . . (10)
We can imagine this as labelling the tiles of a tiling of R in which the tiles are of lengths
1, 2, 3 respectively. If we identify each tile with its right-hand end point, starting with
a tile of type a (length 1) at the origin, then we end up with a sequence of numbers
. . . -26,-24,-21,-18,-17,-15,-12,-9,-8,-6,-3,0,1,3,4,6,9,10,12,13,15,18,19 . . . (11)
The main property of this sequence is its invariance under the transformation
x 7→ 3x. This self-similarity with a rational scaling factor is the signal that there
may be a p-adic interpretation. At the same time, as 3 is not a unit, this sequence is a
candidate for a so-called limit-periodic point set, compare [5]. In fact, this sequence can
be given a 3-adic interpretation. The coordinates of the tiles of the three types (resp.
their right hand endpoints) can be explicitly given as follows:
S∞ k−2
• type a : k=2 (1 + 3 + . . . + 3 ) + 3k Z
S∞ k−2
• type b : k=2 (2 + 1 + 3 + . . . + 3 ) + 3k Z
S∞

• type c : 32 Z ∪ k=3 (−3 − . . . − 3
k−2
) + 3k Z .
p-adic quasicrystals 6

It is easy to see that these sets are invariant under the process of formation of the
tiles (i.e., rule (9)). Furthermore, their densities in the lattice of integers Z are easily
computed to be each equal to 1/6, thus accounting for the entire tiling ((1+2+3)· 61 = 1).
Interpreting these sets p-adically, we see that they are dense subsets of the unions
of open balls formed by replacing 3k Z by 3k Z c3 in each of the sets above. In this way,
we obtain three “windows”, each with compact closure. Furthermore, it is not hard to
see that each of them has a boundary with just finitely many points.
Using the lattice L := {(n, n)|n ∈ Z} ⊆ R × Z c3 we obtain a cut and project
scheme and conclude that the point sets corresponding to each of the three tile types is
a model set. As a consequence, the sequence is an example with pure point diffraction
spectrum. This observation is consistent with the following application of Dekking’s
criterion [4] to a locally equivalent sequence [1], which would share pure pointness with
our above example due to the invariance of this spectral property under mutual local
derivability. Indeed, replacing the configuration ab by a new tile A of length 3 one
obtains a sequence invariant under the substitution rule A → AAc, c → Acc which has
a pure point diffraction spectrum because the new substitution rule is of constant length
and exhibits a so-called coincidence [4].

4. The chair tiling

The 2D chair tiling is defined by the substitution rule in Figure 1a. This has recently
been shown to display a pure point diffraction spectrum [22]. Instead of working with
the chair tiling directly, we introduce a convenient modification by substituting each
“chair” by three decorated squares as in Figure 1b. Since this transformation is local in
the sense of [1] and can be locally inverted by the rule given in Figure 1c, it follows that
showing that the chair tiling is a cut and project tiling is equivalent to showing that the
modified tiling is. With the help of the transformation rules, one sees immediately that
the modified tiling fulfils the substitution rule in Figure 1d.
A particular tiling T0 in the LI class under consideration can be defined in the
following way. Starting with a decorated square S0 of side length 1 centered at the
origin of a fixed coordinate system where the arrow points towards the upper right
corner (1/2, 1/2), one performs successively the following two steps:
(i) perform the affine transformation
1
T : x 7→ T x := 2Rx + (~e1 + ~e2 ), (12)
2
where R denotes rotation by π/2 and ~ei are the canonical unit vectors;
(ii) apply the (appropriately rotated) substitution rule of Figure 1d.
p-adic quasicrystals 7

a) b) c)

d) e) f)

Figure 1. Geometric realization of the chair tiling.

This way, larger and larger portions of a unique member of the LI class are obtained
(see Figure 1e).
As a consequence of the construction, the centers of the decorated squares form the
lattice Z2 . Let Pk be the subset of the centers of squares oriented as Rk S0 , k = 0, 1, 2, 3,
respectively. Clearly, the tiling is completely determined by these subsets.
We will show that each Pk is a model set with internal group G = Z c2 × Z
c2 . As
in our previous one-dimensional example, this will be done by writing Pk as a union of
cosets of certain sublattices of Z2 . Because of the partial symmetry of the decoration of
the second substitution step (see Figure 1f), the tiles which are decorated in this Figure
must repeat with period 4Z2 . If C is the undecorated square underlying S0 , viewed as
a subset of R2 , we must have, as a consequence of the self-similarity involved in the
definition of T0 , that
(Pk ∩ T i C) + 2i · 4Z2 ⊆ Pk (13)
for all i ∈ N, k ∈ {0, 1, 2, 3}. Therefore, if we set Pk,i := Pk ∩ T i C, we get
[ [
Pk = (t + 2i · 4Z2 ) , (14)
i∈N t∈Pk,i

which is the desired decomposition of Pk .

Using the substitution rule, the finite sets Pk,i := Pk ∩T i C can actually be calculated
by recursion:
P0,0 = {0} , Pk,0 = ∅ , k ∈ {1, 2, 3}, (15)
3
[
Pk,i+1 = T i Ml T −i (P(k−nl )4 ,i ) (16)
l=0
p-adic quasicrystals 8

for the integers n0 := 0, n1 := 1, n2 := 2, n3 := 1 and the affine transformations Ml

given by M0 x := x, M1 x := Rx + ~e1 , M2 x := R2 x + ~e1 + ~e2 , M3 x := Rx + ~e2 .
In much the same way as in the one-dimensional example, the decomposition
(14) leads to a description of Pk as model sets. The lattice Z2 is embedded in
G in the canonical fashion. In the embedding space R2 × G we choose the lattice
L := {(~n, ~n)|~n ∈ Z2 }. For each i ∈ N, the closure with respect to the 2-adic topology of
the sublattice 2i · 4Z2 is an open and compact subgroup of G, actually equal to 2i · 4G.
Replacing Z2 by G in (14) and taking the 2-adic closure gives the description of windows
Ωk ⊆ G in G. It is easily seen that Pk is the model set using G as internal space, L as
lattice and Ωk as window.
Finally, we show that the boundary of each open set Ωk has Haar measure 0. Let µ
be the Haar measure on G and assume it is normalized to µ(G) = 1. Then the measure
of any coset t + 2i · 4 · G is 1/[G : (2i · 4 · G)]. On the other hand, the proportion of
points of Z2 lying in the coset t + 2i · 4Z2 is 1/[Z2 : (2i · 4Z2 )] which is exactly the same
number. Then the proportion of points of Z2 occupied by the cosets belonging to Pk
S P
is exactly the same as µ(Ωk ). Since 3k=0 Pk = Z2 we obtain 3k=0 µ(Ωk ) = 1. From
P P
Ω0 ∩ (Ω1 ∪ Ω2 ∪ Ω3 ) = ∅ we have 1 ≥ µ(Ω0 ) + 3k=1 µ(Ωk ) ≥ 3k=0 µ(Ωk ) = 1 and so
µ(Ω0 ) = µ(Ω0 ). Similarly, one gets µ(∂Ωk ) = 0 for k = 0, 1, 2, 3 as required.

5. A limit-quasiperiodic example

The substitution matrix of the primitive two letter substitution system

a → aab b → abab (17)
√
has the Perron-Frobenius eigenvalue λ := 2+ 2 which is a Pisot-Vijayaraghavan number
but not a unit. Therefore, any possible description as a model set of the resulting
substitution sequence will have to use more complicated groups than Rn as embedding
space.
A geometric representation of the substitution system is obtained by replacing
√
symbols a and b by intervals of length ℓ(a) = 1 and ℓ(b) = 2. If we denote the sets of
left endpoints of the a and b intervals by Λa and Λb , the substitution rule leads to the
following system of recursion relations:
Λa = (λΛa ) ∪ (ℓ(a) + λΛa ) ∪ (λΛb ) ∪ (ℓ(a) + ℓ(b) + λΛb ) (18)
Λb = (2ℓ(a) + λΛa ) ∪ (ℓ(a) + λΛb ) ∪ (2ℓ(a) + ℓ(b) + λΛb ) (19)
where the right hand sides represent disjoint unions. Both Λa and Λb are subsets of
√
the group Z[ 2] which can be mapped onto Z2 ⊆ R2 by sending ℓ(a) to ~e1 and
√
ℓ(b) to ~e2 . The transformation t 7→ λt in Z[ 2] induces the linear transformation
φ : ~e1 7→ 2~e1 + ~e2 , ~e2 7→ 2~e1 + 2~e2 of R2 . Notice that φ(Z2 ) ⊂ Z2 , but φ(Z2 ) 6= Z2 .
p-adic quasicrystals 9
√
The transformation φ has the two eigenvalues 2± 2; we may identify the “physical”
√
space R with the subspace V of R2 corresponding to the eigenvalue 2 + 2. Then, all
points Λa and Λb of a substitution sequence according to (17) are images of uniquely
determined points of Z2 under the projection onto the physical space V along the
second invariant subspace of φ. (Note that, as the transformation matrix of φ is not
normal, this projection is not orthogonal with respect to the canonical metric of R2 ;
see Figure 2.) Because the second eigenvalue is smaller than 1, the preimages must
lie in a bounded strip parallel to V , and it is easily calculated that, for the sequence
generated from ba, where the middle vertex is the origin, the preimages lie in the strip
√
V + {t(0, −1 − 2) | 0 ≤ t ≤ 1} (see Figure 2).
The problem is that not all points of Z2 which are in the strip are preimages of
points in the sequence, therefore the embedding so far does not exhibit the sequence
as a model set. However, there is an open substrip whose model set is a subset of
the sequence. This can be seen by observing that the preimages of the sequence can be
connected by a path which only passes along horizontal and vertical bonds in the square
lattice (cf. Figure 2). If one would omit any point of Z2 in the strip
√ √
V + {(0, − 2/2) + t(0, −1 − 2/2) | 0 ≤ t ≤ 1} , (20)
then no such connected path would be possible any more. This observation is the
analogue of finding periodic subsets in the limit periodic examples; all further steps are
more or less completely determined.
←−
We extend R2 by the inverse limit G := Z 2 (φ) (see Eq. (7)) and embed Z2 in
R2 × G in the canonical fashion as the lattice L. The homomorphism φ can be uniquely
extended to R2 × G. From the above considerations, if we take as internal group G′ the
product of the second eigenspace of φ with G, we find an open window in G′ such that
the corresponding model set is a subset of the substitution sequence Λ := Λa ∪ Λb . The
recursion relations (18) and (19) can be transfered to the internal group G′ . This gives
an iterated function system for two windows Ωa and Ωb related to type a and b vertices.
This system has a unique pair of compact sets (Ωa , Ωb ) as its attractor.
Obviously, Λ ⊆ Λ(Ω) for Ω := Ωa ∪ Ωb . Due to the recursion relation (18), λΛ is
a subset of Λa . From the observation (20) we can find an open subset of G inside Ωa .
Then, using the recursion relation both for Λa,b and Ωa,b we can find an open set U ⊂ Ω
which has Ω as its closure such that Λ(U) ⊆ Λ ⊆ Λ(Ω). Since Ω\U has no interior,
Λ(U) and Λ(Ω) differ only on a set of points that is not relatively dense, i.e. on a set
(in R) that has gaps of arbitrary length.
By the argument at the end of the Introduction, we can find a c ∈ G′ so that
Λ(c + Ω) is regular, and hence repetitive, as is the original substitution sequence Λ.
Now let us show that Λ and Λ(c + Ω) are locally isomorphic, thus establishing Λ to be
a model set as defined in the Introduction.
p-adic quasicrystals 10

b
a
b
b a a

b a a

b a
a
b
b a a
a a

Figure 2. The limit quasiperiodic example.

From the above argument, we know that there are arbitrarily long intervals where
Λ(U), Λ and Λ(Ω) coincide. Let us select such an interval of length R. Then, we also
know that there exists a relatively dense set of translations t such that t + Λ(c + Ω)
coincides both with Λ(U) and Λ(Ω) on that interval, too. Since Λ is repetitive, this
establishes that Λ and Λ(c + Ω) are locally isomorphic.
This finally reveals the points Λa and Λb as model sets based on the mixed internal
←−
space R × Z 2 (φ).

6. Diffraction

The diffraction of a generalized model set can be calculated in much the same way as
for model sets in the conventional framework [20]. Given a model set Λ in Rd , one can
show that its characteristic Dirac comb, i.e. the measure
X
ω = ωΛ := δt , (21)
t∈Λ

has a unique autocorrelation,

1
γ = γω := lim (ωΛr ∗ ω̃Λr )
r→∞ vol(Br (0))
1 X
= lim δt−s , (22)
r→∞ vol(Br (0))
s,t∈Λ r

where Br (0) is the ball of radius r around 0, Λr := Λ ∩ Br (0) and ω̃ denotes the measure
defined by (ω̃, φ) = (ω, φ̃) with φ̃(x) := φ(−x).
p-adic quasicrystals 11

A good theory of diffraction exists for model sets under the additional assumption
W2 The boundary of Ω has measure 0 (measure being the Haar measure of G).
If internal space is Euclidean, this is tantamount to saying that the window is a Riemann
measurable set.
If one interprets the measure ω as a set of point scatterers at the sites of Λ, then
the corresponding diffraction pattern is the Fourier transform γ̂ of the autocorrelation
γ. This Fourier transform γ̂ is itself a positive measure and has, for general model sets
Λ with property W2, only a point component, i.e., can be written in the form
X
γ̂ = C(k)δk (23)
k∈F

with non-negative coefficients C(k). The set F in (23) is the projection into Rd of the
dual lattice of L in the dual of Rd × G.
Without going into the detailed calculation, let us give the result in the case of the
limit-periodic substitution sequence (9) for which we can easily verify condition W2 –
indeed, the boundary of the window is a finite point set. We denote the right-hand end
points of the intervals a, b and c by Λa , Λb and Λc , and consider them as the positions
of point scatterers of strengths ha , hb and hc , respectively. The Fourier transform of the
autocorrelation γω of the measure
X X X
ω = ha δt + hb δt + hc δt (24)
t∈Λa t∈Λb t∈Λc

is then given by
X
γ̂ = |ha Aa (k) + hb Ab (k) + hc Ac (k)|2 δk (25)
k∈F

with the “amplitudes”

 
1 πim/3n −πim/3 (−1)m
Aa (k) = n e e + ,
3 2
 
1 −πim/3n−1 −πim/3 (−1)m
Ab (k) = n e e + , (26)
3 2
 
1 −πim/3n−1 πim/3 (−1)m
Ab (k) = n e e + .
3 2

The sum in (25) runs over the Fourier module F ,

n m o
F := k = n | (n = 2, m ∈ Z) or (n ≥ 3, m ≡ / 0 mod(3)) , (27)
3
namely the set of all rational numbers k whose denominators are, at worst, powers of
3. Each such number k is uniquely expressible in the form indicated in (27). It is this
one-to-one parameterization that appears in (26).
p-adic quasicrystals 12

It is easy to see that F is indeed the projection into R of the dual of Z in the dual
of R × Zc3 .
In the case of the chair tiling, we already established W2, and the diffraction
can be calculated along similar lines to the previous example. The limit-quasiperiodic
substitution system of Section 5 is a lot more complicated, and we have not even been
able to verify W2 so far.

7. Comments

The formalism of model sets has been shown to encompass situations not hitherto
considered within its scope by using internal spaces with non-Euclidean topologies.
The situations in which such topologies occur are signalled by the presence of chains of
decreasing sublattices of ever increasing scale.
Let us point out a few more examples. The period doubling substitution rule
a → ba, b → aa, which is known to have pure point spectrum from Dekking’s
criterion [4], gives rise to a 2-adic model set. One of the oldest aperiodic tilings is
the Robinson tiling [18], which is based on a set of six decorated squares. This tiling
has an interpretation in terms lattices of overlapping squares (see [6] for a picture) which
clearly shows its 2-adic nature (something already realized by Robinson). In fact, the
centers of the tiles of each type form a 2-adic model set. In the course of working out
these examples, we discovered that the chair tilings and the Robinson square tilings are
actually closely related. Suitably decorated, the chair tiling can be transformed into a
Robinson tiling and in the reverse direction, suitably undecorated, the Robinson tiling
can be transformed into a chair tiling.
The sphinx tiling as well as the new hexagonal tiling of Penrose [16] undoubtedly
also admit 2-adic interpretations.
The substitution tilings described above have so far been considered as belonging
to the classes of aperiodic tilings called limit-periodic and limit-quasiperiodic tilings,
compare [5]. Potential limit-(quasi-)periodic tilings can be recognized by displaying an
inflation/deflation symmetry in the sense of [1] with an inflation multiplier that is an
algebraic integer larger than 1, but not a unit. Not all of them will display a pure point
diffraction spectrum, as can be seen from the Thue-Morse chain (defined by a → ab,
b → ba) or a variant of our system (9) (defined by a → ab, b → abc, c → ccab). These
cases cannot be model sets. Our analysis shows nevertheless that a unified description of
at least some of the cases with pure point spectrum is possible if one slightly generalizes
the class of internal spaces which are admitted.
This generalization turns out to be a very natural one. Many properties of
conventional model sets can be proved to hold also in the more general case. Among
p-adic quasicrystals 13

these are the uniform densities of general model sets (see [19]) and, if also W2 is fulfilled,
the pure point character of the diffraction spectrum (see [20]). It would be nice to find
an exhaustive criterion for those cases with pure point diffraction spectrum.

Acknowledgments

The authors thank L. Danzer for his interest and helpful discussions on the chair tiling.
RVM is grateful to the Natural Sciences and Engineering Council of Canada for the
continuing support of his research. MS thanks the Pacific Institute of Mathematical
Sciences for support of his research in Canada.

References

[1] M. Baake and M. Schlottmann, “Geometric Aspects of Tilings and Equivalence Concepts”,
in: Proc. of the 5th Int. Conf. on Quasicrystals, eds. C. Janot and R. Mossery, World
Scientific, Singapore (1995), pp. 15–21.
[2] N. Bourbaki, Topology, Vol. 1, Addison-Wesley, Reading (1966).
[3] N. G. de Bruijn, “Algebraic theory of Penrose’s non-periodic tilings of the plane”, part I:
Math. Proc. A84 (1981) 39–52, and part II: Math. Proc. A84 (1981) 53–66.
[4] F. M. Dekking, “The spectrum of dynamical systems arising from substitutions of constant
length”, Z. Wahrscheinlichkeitstheorie 41 (1978) 221–239.
[5] F. Gähler and R. Klitzing, “The diffraction pattern of self-similar tilings”, in: The
Mathematics of Long-Range Aperiodic Order, ed. R. V. Moody, NATO ASI Series C
489, Kluwer, Dordrecht (1997), pp. 141–74.
[6] B. Grünbaum and G. C. Shephard, Tilings and Patterns, Freeman, New York (1987).
[7] E. Hewitt and K. A. Ross, Abstract Harmonic Analysis, Vol. 1, 2nd ed., Springer, New
York (1979).
[8] A. Hof, “Diffraction by aperiodic structures”, in: The Mathematics of Long-Range
Aperiodic Order, ed. R. V. Moody, NATO ASI Series C 489, Kluwer, Dordrecht (1997),
pp. 239–68.
[9] A. Katz and M. Duneau, “Quasiperiodic patterns and icosahedral symmetry”, J. Physique
47 (1986) 181–96.
[10] P. Kramer, “Non-periodic central space filling with icosahedral symmetry using copies of
seven elementary cells”, Acta Cryst. A38 (1982) 257–64.
[11] P. Kramer and R. Neri, “On periodic and non-periodic space fillings of Em obtained by
projection”, Acta Cryst. A40 (1984) 580–7, and Acta Cryst. A41 (1985) 619 (Erratum).
[12] P. McMullen, “Duality, sections and projections of certain Euclidean tilings”, Geom.
Dedicata 49 (1994) 183–202.
[13] Y. Meyer, Algebraic Numbers and Harmonic Analysis, North-Holland, Amsterdam (1972).
[14] R. V. Moody, “Meyer sets and their duals”, in: The Mathematics of Long-Range Aperiodic
Order, ed. R. V. Moody, NATO ASI Series C 489, Kluwer, Dordrecht (1997), pp. 403–41.
p-adic quasicrystals 14

[15] J. Neukirch, “The p-adic numbers”, in: Numbers, eds. H.-D. Ebbinghaus et al., Springer,
New York (1990), pp. 155–178.
[16] R. Penrose, “Remarks on tiling: Details of a (1+ǫ+ǫ2 )-aperiodic set”, in: The Mathematics
of Long-Range Aperiodic Order, ed. R. V. Moody, NATO ASI Series C 489, Kluwer,
Dordrecht (1997), pp. 467–97.
[17] M. Queffelec, “Spectral study of automatic and substitutive sequences”, in: Beyond
Quasicrystals, ed. F. Axel and D. Gratias, Springer, Berlin (1994), pp. 369–414.
[18] R. M. Robinson, “Undecidability and nonperiodicity of tilings of the plane”, Inv. Math.
44 (1971) 177–209.
[19] M. Schlottmann, “Cut-and-project sets in locally compact Abelian groups”, in: Quasi-
crystals and Discrete Geometry, ed. J. Patera, Fields Institute Monographs, Vol. 10,
AMS, Rhode Island (1998), pp. 247–64.
[20] M. Schlottmann, “Generalized model sets and dynamical systems”, to appear in:
Directions in Mathematical Quasicrystals, eds. M. Baake and R. V. Moody, CRM
monograph series, AMS, Rhode Island (1998), in preparation.
[21] J.-P. Serre, A Course in Arithmetic, Springer, New York (1973).
[22] B. Solomyak, “Dynamics of self-similar tilings”, Ergod. Th. & Dynam. Syst. 17 (1997)
695–738.