Lecture Series on Lyapunov Exponents

Neil Mañibo
Contents

1 Basic Notions 2
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Lyapunov Exponents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Filtrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Regularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Matrix cocycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Main Theorems and Central Results 8

3 Lyapunov Exponents in Diffraction Theory of Aperiodic Structures 10

3.1 Aperiodic structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2 Diffraction of aperiodic structures . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3 Inflation tilings and Fourier cocycles . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.4 Absence of absolutely continuous diffraction . . . . . . . . . . . . . . . . . . . . . 15
3.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.5.1 Bijective Abelian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.5.2 Fibonacci Tiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Bibliography 19

1
1 Basic Notions

1.1 Introduction
Let us begin with an easy problem. Let M = ( 11 10 ). The
√
spectrum of M consisting of its
−1 1+ 5

eigenvalues is given by σ(M ) = τ, −τ , where τ = 2 is the golden ratio. We now have
the following question.

Question 1.1.1. What is the value of

1
χ = lim log kM n k?
n→∞ n
Intuitively, one guesses that the answer must be χ = log(τ ), with τ being the spectral radius
of M , which dictates the dominant exponential behavior of M n . One can even prove this easily
since M is diagonalisable and one can instead consider the product of the diagonal matrices
containing the eigenvalues.
Now, let us make it slightly difficult by looking at two matrices instead of one. Consider
the matrices M1 = ( 21 11 ) and M2 = ( 32 11 ) , whose respective spectra are given by σ(M1 ) =
√
τ , −τ −2 and σ(M2 ) = 2 ± 3 . Define M (n) to be a random product (say both matrices
 2 

have equal probability p = 21 ) of n matrices from the set M1 , M2 , which one can formally


write as
M (n) = Min−1 Min−2 · · · Mi1 Mi0

where ij ∈ 1, 2 .

Question 1.1.2. What is the generic value of

1
χ = lim log kM (n) k?
n→∞ n

Does it even exist?

Indeed, this limit exists for a.e. sequence (ij )j∈N0 . One might guess that it is just given by
the average of log(λ1 ) and log(λ2 ). Surprisingly, this is not the case, and it has been calculated
that
1 √

χ = 1.143311035 6= log(τ 2 ) + log(2 + 3) ≈ 1.13969077.
2
The complete derivation for this example, and for a general family of matrices can be found
in [Pol10]. In Kingman’s seminal paper [Kin73], he said that “pride of place among the unsolved
problems of subadditive ergodic theory must go to the calculation of the number χ”, which
gives a natural answer to the question, “why are Lyapunov exponents interesting?”; they are
interesting because they are hard to compute and there is no general way of calculating them.

2
1.2 Lyapunov Exponents
For the following review of basic material, we use [Via13] and [BP07] as our main reference
texts.

Definition 1.2.1. Given a sequence {Mj }j>0 of matrices in Mat(d, C), satisfying the condition

1
lim sup log kM (m) k < ∞, (1.1)
m→∞ m

one can consider its Lyapunov exponent χ : Cd → R ∪ {−∞} defined by

1
χ(v) = lim sup log kM (n) vk, (1.2)
n→∞ n

where we have set M (n) := Mn−1 Mn−2 · · · M1 M0 .

The function χ satisfies the following properties.

(1) χ(αv) = χ(v), for all v ∈ Cd , α ∈ R \ 0



(2) χ(v + w) 6 max χ(v), χ(w) , for all v, w ∈ Cd



(3) χ(0) = −∞.

Here, we follow the convention that log(0) = −∞. We also note that χ(v) does not depend on
the norm k · k chosen as they are all equivalent in the finite-dimensional case.

Remark 1.2.2. The property given in Eq. 1.1 ensures that χ(v) < ∞. This is immediately
satisfied when supj kMj k < ∞ or when limn→∞ n1 log+ kMj k < ∞. ♦

Example 1.2.3. Let X be an r-dimensional smooth submanifold of Rd and g : X → X be

a C 1 -diffeomorphism of X. Let x ∈ X, Tx X be the tangent space at x, and v ∈ Tx X. The
Lyapunov exponent associated to x is given by
1
χ(x, v) = lim log kDg n (x)vk,
n→∞ n
where Dg n (x) is the Jacobi matrix of g n at x. ♦

Example 1.2.4 ( [Bar17, Ex. 1.1.2]). Let A : R>0 → Mat(d, R). The Lyapunov exponent χ
associated with the differential equation v 0 = A(t)v is given by
1
χ(v0 ) = lim sup log kv(t)k
t→∞ t

where v(t) is the solution of the IVP

v 0 = A(t)v and v(0) = v0 .

3
1.3 Properties
Proposition 1.3.1. Let χ be the Lyapunov exponent defined above.

(1) If χ(v) 6= χ(w), then χ(v + w) = max χ(v), χ(w)

(2) If v1 , . . . , vm ∈ Cd , α1 , . . . , αm ∈ R \ 0 , then



χ(α1 v1 + · · · + αm vm ) = max χ(vi ) | 1 6 i 6 m .

(3) If for some v1 , . . . , vm ∈ Cd , the values χ(v1 ), . . . , χ(vm ) are distinct, then v1 , . . . , vm are
linearly independent.

(4) χ attains at most d distinct finite values

Proof.

(1) Suppose χ(v) < χ(w). One then has


χ(v + w) 6 χ(w) = χ(v + w − v) 6 max χ(v + w), χ(v)

If χ(v + w) < χ(v), then χ(w) 6 χ(v), which is a contradiction. Hence, χ(v + w) > χ(v),

implying χ(v + w) = χ(w) = max χ(v), χ(w) .

(2) Follows from Properties 1 and 2, and (1).

(3) Assume on the contrary that v1 , . . . , vm are linearly dependent, i.e.,

α1 v1 + · · · + αm vm = 0,

with not all αi = 0, while χ(v1 ), . . . , χ(vm ) are distinct. This yields

χ(α1 v1 + · · · + αm vm ) = −∞ = max χ(vi ) | 1 6 i 6 m, 6= −∞,

which is a contradiction.

(4) Follows from (3)

1.4 Filtrations
From the distinct values χ1 , . . . , χd0 , one can construct a filtration of Cd , i.e., a nested sequence
0
of subspaces {V}di=1
0
Cd = V 1 ) V 2 ) . . . ) V d 6= {0} (1.3)

such that χ(v) = χi , for all v ∈ V i \ V i+1 .

Remark 1.4.1. In the case where {Mj }j>0 is made up of a single matrix M , the Lyapunov
exponents χi are given by log |λi |, where λi are the eigenvalues of M and, V i \ V i+1 are the
corresponding (possibly generalised) eigenspaces. When {Mj }j>0 is a convergent sequence with
limit M , the values of the exponents are also determined by the eigenvalues of M . ♦

4
1.5 Regularity
Definition 1.5.1. A sequence {Mj }j>0 is said to be forward regular if

d
1
 X 0
log  det M (n)  =

lim χi , (1.4)
n→∞ n
i=1

provided that the limit exists.

Here, χ01 > . . . > χ0d are the values attained by χ, counted with their multiplicities. Mere
existence of the limit does not guarantee forward regularity, as we shall see in the following
example.
! !
1 0 1 0
Example 1.5.2. Let M0 = and Mj = , for each j > 1, which makes
2 4 −2j+1 4
!
1 0
M (n) = . A direct calculation gives det M (n) = 4n , which yields
2n 4n

1
log  det M (n)  = 2 log(2)


lim
n→∞ n

However, it is easy to check that for v1 = (1, 0)T and v2 = (0, 1)T , one has

χ(v1 ) = log(2) and χ(v2 ) = 2 log(2),

and hence
1
log  det M (n)  = 2 log(2) 6= χ1 + χ2 = 3 log(2).


lim
n→∞ n
♦

1.6 Matrix cocycles

Let (X, T ) be a topological dynamical system. A measure µ on X is invariant with respect to
T if µ(E) = µ(T −1 (E)), for all Borel sets E. With the invariant measure µ, (X, T, µ) becomes a
measure-theoretic (or metric) dynamical system.
A transformation T is ergodic with respect to µ, if T −1 (E) = E implies either µ(E) = 0 or
µ(E) = 1. Analogously, µ is an ergodic measure for the topological dynamical system (X, T ) if
the same condition is satisfied.

Example 1.6.1 (Irrational rotations). Let X = [0, 1) ' S 1 and T be a rotation given by
Tα : x + α, for some α ∈ [0, 1). The map Tα is ergodic with respect to the Haar measure on X
(which is the Lebesgue measure) if and only if α is irrational.

Theorem 1.6.2 (Birkhoff’s Ergodic Theorem). Let (X, T, µ) be an ergodic measure-preserving

dynamical system, g ∈ L1 (µ). Then, one has
n−1
1X
Z
lim g(T j x) = g(ξ)dµ(ξ),
n→∞ n X
j=0

for µ-a.e. x ∈ X.

5
 One way to generate sequences of matrices is via cocycles. Consider a measure-preserving
dynamical system (X, f, µ) and a measurable matrix-valued map A : X → Mat(d, C).

Definition 1.6.3. A skew linear map F : X × Cd → X × Cd defined by (x, v) 7→ f (x), A(x)v

is called a linear cocycle over f , where f is the base dynamics of the cocycle.

We call F ergodic over (X, f, µ) if f is ergodic with respect to µ. An iteration of this function
yields the pair F n (x, v) = (f n (x), A(n) (x)v), where the induced fibre action on Cd is determined
by the matrix product

A(n) (x) = A(f n−1 (x)) · . . . · A(f (x))A(x).

Unless otherwise stated, we assume the base dynamics to be fixed, and we refer to A(n) (k) as
the matrix cocycle.

Example 1.6.4.

(1) Let Ω ⊂ Mat(d, C) be compact. Let X = ΩZ with the (left-sided) shift operator S on
X, with (Sx)k = xk+1 , for {xk }k∈Z ∈ ΩZ , and µ a probability measure on Ω. Consider
the locally constant map A : x = (. . . , x−1 , x0 , x1 , . . .) 7→ A(x0 ). Then, (S, A) defines a
cocycle over X × Cd . Furthermore, S is ergodic with respect to the product measure µZ .

(2) Let X = Td := Rd /Zd , A : Td → Mat(d, C), and M f be a toral endomorphism given

by Mf : x 7→ (M x) mod 1, where M ∈ Mat(d, Z). It is well known that M f is ergodic
with respect to Lebesgue measure whenever det M 6= 0 and M does not have eigenvalues
which are roots of unity [EW11, Cor. 2.20], and is invertible whenever M ∈ GL(d, Z), i.e.,
det M = ±1. As in the first example, (Mf, A) defines a matrix cocycle. ♦

For sequences arising from cocycles, more specific versions of Eq. (1.2) and Eq. (1.3) for the
Lyapunov exponent χ : Cd × X → R ∪ {−∞} and the x-dependent filtration it defines read
1 0
χ(v, x) = lim sup log kA(n) (x)vk and Cd = Vx1 ) Vx2 ) . . . ) Vxd (x) 6= {0} ,
n→∞ n

with χ(v, x) = χi (x), for all v ∈ Vxi \ Vxi+1 . We say that A(n) (x) at a given point x is forward
regular if the sequence {A(f n (x))}n>0 is forward regular.

Lemma 1.6.5. Let v ∈ Cd \ {0} , x ∈ X. Assuming A(n) (x)−1 exists, one has,

χmin (x) 6 χ(x, v) 6 χmax (x),

where
1 1
χmax (x) = lim sup log kA(n) (x)k and χmin (x) = lim inf log kA(n) (x)−1 k−1 .
n→∞ n n→∞ n
Proof. Note that the following holds for all non-zero v,

kA(n) (x)−1 k−1 kvk 6 kA(n) (x)vk 6 kA(n) (x)kkvk.

The claim then directly follows by taking the logarithm, and the lim sup and the lim inf of the
upper and the lower bound, respectively.

6
 Define φ+ (x) := max {0, φ(x)}. The following result on the extremal exponents is due to
Furstenberg and Kesten [FK60]; see also [Via13, Thm. 3.12].

Theorem 1.6.6 (Furstenberg–Kesten). Let F : X × Rd → X × Rd be a matrix cocycle defined

by F (x, v) = (f (x), A(x)v), where A : X → GL(d, R) is measurable, and X is compact. If
log+ kA±1 k ∈ L1 (µ), the extremal exponents χmin (x) and χmax (x) exist as limits for a.e. x ∈ X.
Moreover, these functions are f -invariant and are µ-integrable.

7
2 Main Theorems and Central Results

The following generalisation of Birkhoff’s ergodic theorem for subadditive functions is due to
Kingman [Kin73]; compare [Via13, Thm. 3.3].

Theorem 2.0.1 (Kingman’s subadditive ergodic theorem). Let X be a compact space and
assume f : X → X to be invariant with respect to µ. Let {φn } be a sequence of functions such
that φ+ 1
1 ∈ L (µ) and

φm+n 6 φm + φn ◦ f m holds for all m, n > 1. (2.1)

Then, ( φnn ) converges µ-a.e. to some φ : X → [−∞, ∞). Moreover, φ+ is integrable and
1 1
Z Z Z
φ(ξ)dµ(ξ) = lim φn (ξ)dµ(ξ) = inf φn (ξ)dµ(ξ) ∈ [−∞, ∞).
X n→∞ n X n n X

Furthermore, when f is ergodic, φ is constant for µ-a.e. x ∈ X.

Example 2.0.2. A relevant class of functions satisfying Eq. 2.1 are those arising from cocycles,
which are of the form φm = log kA(m) (x)k, where k · k is a submultiplicative matrix norm. One
can explicitly see this by looking at the decomposition of A(m+n) given by

A(m+n) (x) = A(f m+n−1 x) · · · A(f m x) · A(f m−1 x) · · · A(f x)A(x) .

| {z } | {z }
A(n) (f m x) A(m) (x)

Example 2.0.3. Not all matrix norms are submultiplicative. An example of a nonsubmulti-
plicative norm is
kM kmax = max |mij |
i,j

One can easily check for M = ( 11 10 ) that this is not submultiplicative, i.e.,

kM 2 kmax = 2 > 1 = kM kmax kM kmax .

For ergodic real-valued cocycles, one has the following central result in the theory of Lyapunov
exponents, which is due to Oseledec [Ose68]; see [Via13, Thm. 4.1] and [BP07, Thm. 3.4.3].

Theorem 2.0.4 (Oseledec’s multiplicative ergodic theorem, one-sided). Let f be an ergodic

transformation on a compact space X with an invariant measure µ. Let A : X → GL(d, R)
be measurable, such that the condition log+ kAk ∈ L1 (µ) holds. Then, for µ-a.e. x ∈ X, the
cocycle A(n) (x) is forward regular. Moreover, for these x, the Lyapunov exponents χi (x) are
constant, i.e., there exist real numbers χ1 , . . . , χd0 , and a filtration
0
Rd = Vx1 ) Vx2 ) . . . ) Vxd 6= {0}

8
such that
1
lim log kA(n) (x)vi k = χi
n→∞ n

for all vi ∈ Vxi \ Vxi+1 .

There exists an even stronger notion of regularity, also known as Lyapunov–Perron regularity.
This requires both the matrix-valued function A and the map f to be invertible so that one can
define A(n) (x), for n < 0. Under these invertibility assumptions, and that log+ kA−1 k ∈ L1 (µ),
one gets the following two-sided version of Theorem 2.0.4.

Theorem 2.0.5 (Oseledec’s multiplicative ergodic theorem, two-sided). Let f be an invertible

ergodic transformation on a compact space X with an invariant measure µ. Let A : X →
GL(d, R) be measurable, such that the condition log+ kA±1 k ∈ L1 (µ) holds. Then, for µ-a.e.
x ∈ X, the cocycle A(n) (x) is Lyapunov–Perron regular. Moreover, for these x, the Lyapunov
exponents χi (x) are constant, i.e., there exist real numbers χ1 , . . . , χd0 , and a splitting
0
Rd = Ex1 ⊕ Ex2 ⊕ . . . ⊕ Exd 6= {0}

such that
1
lim log kA(n) (x)vi k = χi
n→±∞ n

for all vi ∈ Exi . Furthermore, the subspaces Exj are invariant with respect to A(x), i.e., for a.e.
x, A(x)Exi = Efi (x) .

9
3 Lyapunov Exponents in Diffraction Theory of
Aperiodic Structures

3.1 Aperiodic structures

Before 1992, the definition of a crystal by the IUCr is a material with a periodic structure,
which mathematically corresponds to a lattice in Rn (where n is usually 2 or 3). This has
since been changed to those solids which display pure point diffraction. The theory of aperiodic
order deals with structures having no translational symmetry but might possess a certain type
of long-range order. In physics, this long-range order is usually characterised by a set of points
of high intensity in the diffraction image obtained when one subjects the said structure to
a diffraction experiment. These bright points are known as Bragg peaks, and their presence
implies that the structure has pure point diffraction component. This area of research was
boosted by Schechtman’s discovery [SBGC84] of a material with fivefold rotational symmetry,
and hence does not possess a lattice structure (could not be indexed using any of the 14 Bravais
lattices in three dimensions), but has pure point diffraction; see Fig. 1.

Figure 1: Diffraction image of the AlMn-alloy quasicrystal discovered by Schechtman, taken

from [SBGC84]

It is an easy exercise to show that in R2 and R3 , lattices only admit a limited number of
allowable rotational symmetries.

Proposition 3.1.1 ( [BG13, Lem. 3.2]). Let Γ be a lattice in Rd , d ∈ {2, 3} and R ∈ O(d). If
RΓ = Γ , then the characteristic polynomial P (λ) = det(R − λI) ∈ Z[λ].

10
 
Proof. Let b1 , . . . , bd be a basis for Γ , so that Γ = Zb1 ⊕ · · · ⊕ Zbd . If RΓ = Γ , one then has
X
Rbi = bj aji , with aji ∈ Z.
j

In matrix form, one has RB = BA, where B is the basis matrix of Γ . Since B is invertible, one
gets
R = BAB −1 ,

where A ∈ GL(d, Z). The claim follows since A and R share the same characteristic polynomial.

Figure 2: Illustration of the imposibility of five-fold rotational symmetry for lattices

Corollary 3.1.2 ( [BG13, Cor. 3.1]). A lattice Γ ⊂ Rd with d ∈ 2, 3 can have n-fold rotational


symmetry only for n ∈ 1, 2, 3, 4, 6 .

Proof. In dimension d = 2, a rotation matrix is given by

!
cos(θ) − sin(θ)
Rθ = ,
sin(θ) cos(θ)

whose characteristic polynomial is P (λ) = λ2 − 2 cos(θ) + 1. If P (λ) ∈ Z[λ], then one has
| cos(θ)| ∈ 0, 21 , 1 , which immediately implies the claim.


For d = 3, one uses the well-known fact that Rφ is similar to a matrix of the form
 
cos(φ) − sin(φ) 0
Rφ0 =  sin(φ) cos(φ) 0 ,
 

0 0 1

whose characteristic polynomial is P (λ) = (1 − λ)(λ2 − 2 cos(θ) + 1), which yields the same
conclusion.

Before Schechtman’s discovery, the first paradigm shift from purely periodic structures stemmed
from the study of almost periodic functions; see [Boh93, Esc04, Boh47, Bes54]. Notably, notions
of Fourier transformability for unbounded measures had already seen reasonable progress by
the early ’70s; see [AdL74].
Alongside these developments in harmonic analysis was a proliferation of important results
on non-periodic tilings. The undecidability of the domino problem was established by Berger
in 1966 [Ber66], which meant a tiling of the plane via a finite set of decorated tiles need not be
periodic.

11
 (a) The Jeandel-Rao Wang tiles (b) A patch of a Jeandel-Rao
tiling

Within a decade, Penrose solved a related but geometrically different problem in his monu-
mental discovery of tilings of R2 by six prototiles having no translational symmetry (and hence
are non-crystallographic) [Pen74].
Although there were lots of works which already applied Fourier analysis on aperiodic tilings,
it was the work of Dworkin [Dwo93] and Hof [Hof95] that set the stage for mathematical
diffraction of aperiodic structures. Under this formalism, one normally views a vertex set Λ
of an aperiodic tiling T as a model for a quasicrystal; see Fig. 3. One distinguishes different
atoms by placing different weights signifying distinct scattering strengths. The non-periodicity
of such tilings imply that one must deal with (weighted) unbounded measures to describe atomic
positions.

(a) A patch of an aperiodic tiling (b) Point sets via vertex points

Figure 3: Aperiodic point sets via tilings

3.2 Diffraction of aperiodic structures

Motto: Mathematical diffraction is Fourier analysis on point sets.
A nice and detailed survey of mathematical diffraction can be found in [BG13, Ch. 9], and a
list of works on Lyapunov exponents in the diffraction setting would contain [BFGR19,BGäM18,
BGrM18, Man17a, Man19].

12
 F
1

0.5

0.4
0.5

0.3

0.2

0.1

0
x
0 5 10 15 20 0 0.5 1

(a) γ
bpp (b) γ
bac (c) γ
bsc

Figure 4: Illustration of the three different spectral components

A point set Λ in Rd is a countable union of singletons. We usually require that these point
sets ares

UD Uniformly discrete: there exists an Rc > 0 such that, BRc (x) ∩ Λ = x for all x ∈ Λ

RD Relatively dense: there exists an Rp > 0 such that BRp (x) contains at least two points in
Λ, for all x ∈ Λ.

Point sets satisfying these two properties are called Delone sets. From an infinite Delone set
Λ ⊂ Rd , one can construct a weighted Dirac comb ω = x∈Λ w(x)δx , which represents the
P

atomic configuration, and distinguishes positions occupied by different atom types by different
values (typically complex) of the weight function w(x).
Aperiodic tilings naturally give rise to these point sets when one considers their vertex sets.
This choice is not unique, as one could also choose a point in the interior of the tile to be the
tile’s control point.
From ω, one then computes the autocorrelation measure γ via γ = ω ~ ω e . Then, one takes the
Fourier transform of γ to get the diffraction measure γ b. The diffraction γ
b is a positive measure,
and hence has the Lebesgue decomposition given by

γ
b=γ
bpp + γ
bac + γ
bsc .

Here, (b
γ )pp is the pure point component and is the analytic analogue of Bragg peaks in a
diffraction experiment, (bγ )ac is the absolutely continuous component represented by a locally-
integrable function h(k) whose non-triviality is usually attributed to a certain level of random-
ness, and in particular, represents what is called diffuse diffraction, and (bγ )sc is the singular
continuous component, which lives on an uncountable set of measure zero and is difficult to
detect in experiments.
One of the main goals of the mathematical diffraction theory is to be able to deduce implica-
tions of properties of the underlying point set Λ to the qualitative properties, such as presence,
absence, density of the support, etc., of each of these three components.

13
3.3 Inflation tilings and Fourier cocycles
We focus on tilings that are generated by inflation rules, which are rules that act on a finite
number of starting tiles called prototiles. The rule % sends a prototile t into a finite non-
overlapping gapless union of the different prototiles and their translates.
In dimension one, the prototiles are intervals in R, and % sends an interval into a finite
concatenation of intervals; see Fig. 5.

Figure 5: The squared Fibonacci inflation rule %2F

The corresponding lengths of the prototiles are |tred | = τ and |tblue | = 1, where τ is the golden
ratio. The inflation rule %2F sends in Fig. 5 sends each tile to a concatenation of tiles, whose total
length is λ = τ 2 times the original length of the starting prototile, hence the term inflation rule.
We call λ the inflation multiplier associated to %2F . Dimension-one inflation rules are normally
derived from symbolic objects called substitutions, where one requires these substitutions to be
primitive, i.e., that they have primitive substitution matrices.
Moreover, this example also admits a bi-infinite tiling fixed point, which one can obtain by ap-
plying the rule infinitely many times to two red tiles joined together. We call this concatenation
that generates the tiling of R a seed. The generated fixed point is given in Fig. 6.

Figure 6: The tiling fixed point T generated by %2F

From this tiling of R, one can then get a point set Λ by collapsing each interval into a single
point, which we situate at the tile’s left endpoint. This yields a point set composed of red and
blue points, which is usually called a coloured point set; see Fig. 7. It is easy to see that since
we started with finitely many prototiles of possibly varying lengths, the resulting point set is
Delone.

Figure 7: The point set Λ derived from T

To this point set, one can then get the measure ω which represents the atomic distribution,
and the measure we need for diffraction analysis. It follows from general theory that this point
set, as well as the tiling it came from, is aperiodic, but one can also show that it has pure point
diffraction.
In higher dimensions, everything works analogously as in one dimension, except that there
is the obvious additional freedom for the geometry of the prototiles. Moreover, the concept of
an inflation multiplier no longer makes sense, hence we normally talk about a linear expansive

14
map Q which satisfies
n
[
Q(ti ) 7−→ tj + Fij
%
j=1
where Fij is a finite set. A two-dimensional inflation rule is given in Fig. 8. For this example,
√
one has Q = λ2 I2 , where λ = 12 (5 + 5).

Figure 8: The inflation rule %GLB which generates the Godrèche–Lançon–Billard tiling in Fig. 3

To an inflation rule %, both in one and in higher dimensions, one can associate an na × na -
matrix which encodes the positions of prototiles ti in supertiles %(tj ). This matrix is called the
Fourier matrix B(k) associated to %. Recall that the positions of tiles are always given by their
left endpoints. These positions can be seen explicitly in B(k).

Figure 9: The squared Fibonacci inflation rule %2F with the corresponding position of tiles in
supertiles

The corresponding Fourier matrix for %2F is given by

!
1 + e2πi(τ +1)k 1
B(k) = ,
e2πiτ k e2πiτ k
where k is a real parameter. For %GLB , one has ten prototiles (up to translation and identifi-
cation), hence the matrix B(k) is a 10 × 10 matrix, where k ∈ R2 . The number of tiles gives
the dimension of B(k), whereas the dimension of the space where the tiling lives dictates the
dimension of k.

3.4 Absence of absolutely continuous diffraction

Let B(k) be the Fourier matrix associated to % and let λ be the inflation multiplier. Consider
the matrix cocycle whose generator is B(k) with base dynamics given by f : k 7→ λk. The nth
level cocycle is given by
B (n) (k) = B(k)B(λk) · · · B(λn−1 k).
Consider the extremal Lyapunov exponent given by
1
χB (k) := lim sup log kB (n) (k)k,
n→∞ n

15
which depends on the real parameter k.
The following main result gives a sufficient criterion for ruling out absolutely continuous
diffraction using the Lyapunov exponent χB (k).

Theorem 3.4.1 ( [BGäM18], 2019). Let % be a one-dimensional inflation rule, with inflation
multiplier λ, and let B(k) be the corresponding Fourier matrix. Under some mild assumptions,
if there is an ε > 0 such that
√
χB (k) 6 log λ − ε
for Lebesgue-a.e. k ∈ R, then the diffraction γ
b of any point set arising from % does not contain
an absolutely continuous component.

Sketch of proof.

(1) Fact: h(k) cannot grow exponentially as k → ∞

(2) From finiteness of prototiles: h(k) is determined by some vectors v(k) ∈ Cna

(3) From the inflation structure: v(k) satisfies some renormalisation equation involving
B(k)

(4) If all Lyapunov exponents for B (n) (k) are positive, then all possible candidates for v(k)
also grow exponentially under B (n) (k)

(5) v(k) = 0 Lebesgue-a.e. =⇒ h(k) = 0 Lebesgue a.e. =⇒ γ

bac = 0

For higher-dimensional examples, we have the following generalisation.

Theorem 3.4.2 ( [BGäM18], 2019). Let % be an inflation rule in Rd , with expansive linear map
Q, and let B(k) be the corresponding Fourier matrix. Under some mild assumptions, if there is
an ε > 0 such that
p
χB (k) 6 log | det Q| − ε
for Lebesgue-a.e. k ∈ Rd , then the diffraction γ
b of any point set arising from % does not contain
an absolutely continuous component.

3.5 Examples
3.5.1 Bijective Abelian
The following family of examples is systematically treated in [BGäM18, Sec. 4].

Example 3.5.1. Let A = {0, 1, 2} and consider %ab to be




 0 7→ 0 2 1 1 2

% : 1 7→ 1 0 2 2 0 .


 2 7→ 2 1 0 0 1


This is an example of a bijective substitution. The zeroth and the third column are C0 = (0, 1, 2)T
and C3 = (1, 2, 0)T , respectively.

16
 Constant-length substitutions are the simplest substitutions which give rise to inflation rules
in the sense that, the inflation multiplier λ is given by the length L, and each letter gets identified
to a tile of unit length. Since tile positions correspond to positions of letters in substituted words,
it suffices to consider the symbolic picture in this setting. The corresponding inflation is given
in Figure 10.

Figure 10: The inflation rule corresponding to %ab

Remark 3.5.2. One can consider the columns of % as permutations in S|A| via

σC` (0, 1, 2, . . . , |A| − 1)T = C` .

Definition 3.5.3. A bijective substitution % is Abelian if the group G “generated by the


columns”, i.e., G = σC` is Abelian.

Proposition 3.5.4. Let % be Abelian. Then

L−1
X
B(k) = e2πi`k D`
`=0

where {D` } are commutative permutation matrices.

Definition 3.5.5. Given a polynomial P (x) ∈ C[x], its logarithmic Mahler measure is given by
the mean Z 1
m(P ) = log |P ( e2πit )|dt.
0

Proposition 3.5.6. Let % be Abelian. Then the |A| Lyapunov exponents of B(k) are given by
χj (k) = m(Pj ), for a.e. k ∈ R, where m(Pj ) is the logarithmic Mahler measure of the polynomial

L−1
X
Pj (x) = ρj (σC` )x` , where ρj is the jth irrep of G.
`=0

Theorem 3.5.7. Let % be a primitive, bijective, Abelian substitution. Then, for a.e. k ∈ R,
√
all Lyapunov exponents of B(k) are strictly less than log λ. Moreover, the corresponding
diffraction γ
b does not contain an absolutely continuous component.

3.5.2 Fibonacci Tiling

The example given by the squared Fibonacci inflation in Fig. 5 is one of the classic examples
of one-dimensional inflations with irrational inflation multiplier λ. For such rules, one cannot
reduce the cocycle B (n) (k) into a cocycle on the torus [0, 1), unlike the previous example where
λ is always an integer because of the constant-length property. Precisely because of this, one
cannot directly apply Oseledec’s theorem to obtain existence and almost everywhere constancy

17
of the exponent χB (k). Nevertheless, due to the quasiperiodicity of B (n) (k) (finiteness of the
fundamental frequencies in the entries of B(k)), one can express B(k) as a section of a periodic
function over Tr , where r√ is the algebraic degree of λ; compare with [BGrM18].
For %2F , λ = τ 2 = 3+2 5 , whose algebraic degree is r = 2. For this inflation, one has the
representation of the Fourier matrix as

B(k) = B(x,
e y)|x=k, y=λk .

This allows one to consider a cocycle on Tr and use this cocycle to bound χB (k). In particular,
one has the following results for general one-dimensional inflations.

Proposition 3.5.8. Let % be a primitive one-dimensional inflation. Under some mild assump-
e on Tr , such that
tions, there exist an ergodic cocycle B

B (n) (k) = B
e (n) (x1 , . . . , xr )|x =k, x =θ k,...,x =θ k
1 2 1 r r−1


where the set 1, θ1 , . . . , θr−1 consists of rationally independent frequencies.

Using some results on averaging almost periodic functions along uniformly distributed se-
quences in [BHL17], one gets the following useful result.

Proposition 3.5.9 (Sequence of uniform upper bound of χB (k)). Let % be a primitive one-
dimensional inflation. Under some mild assumptions,
1
Z
B e (N ) (x)kdµ (x),
χ (k) 6 log kB H
N Tr

e (N ) (x) is an ergodic cocycle on Tr .

for a.e. k ∈ R for all N > 1. Here, B

The previous proposition provides a sequence of uniform upper bounds for the Lyapunov
exponent χB , which one can compute using numerical methods (quasi-Monte Carlo integration).
√
The values for the bounds are then to be compared with log λ, and if for a particular N , the
√
said upper bound is strictly less than log λ, one can then invoke Theorem 3.4.1 to rule out
the presence of absolutely continuous diffraction.
For the squared Fibonacci, the pertinent numerical values are given in Table 3.1.

N 1 2 3 4
1 e (N ) (k)k2
R
N T 2 log kB F 1.5668 1.1091 0.8776 0.7409

Table 3.1: Numerical values for upper bounds for 2χB (k) for %2F . Here k · kF stands for the
Frobenius norm.

At N = 3, the upper bound for 2χB (k) crosses the threshold log λ = 0.9624, which implies that
the point set coming from %2F has no absolutely continuous diffraction. Using exactly the same
method, one can prove that the diffraction of the vertex set of a GLB tiling generated generated
by the inflation rule %GLB given in Figure 8 has essentially singular continuous diffraction;
see [BGäM18, Man19].

18
Bibliography

[AdL74] Argabright L N and de Lamadrid J G, Fourier Analysis of Unbounded Measures

on Locally Compact Abelian Groups, Memoirs Amer. Math. Soc. 145 (1974).

[BFGR19] Baake M, Frank N P, Grimm U and Robinson E A, Geometric properties of a

binary non-Pisot inflation and absence of absolutely continuous diffraction, Studia
Math. 247 (2019) 109–154, arXiv:1706.03976.

[BG16] Baake M and Gähler F, Pair correlations of aperiodic inflation rules via
renormalisation: some interesting examples, Topol. & Appl. 205 (2016) 4–27,
arXiv:1511.00885.

[BGäM19] Baake M, Gähler F and Mañibo N, Renormalisation of pair correlation mea-

sures for primitive inflation rules and absence of absolutely continuous diffraction,
Commun. Math. Phys., in press, arXiv:1805.09650.

[BG13] Baake M and Grimm U, Aperiodic Order, Volume 1: A Mathematical Invitation,

Cambridge University Press, Cambridge (2013).

[BGrM18] Baake M, Grimm U and Mañibo N, Spectral analysis of a family of binary infla-
tion rules, Lett. Math. Phys. 108 (2018) 1783–1805, arXiv:1709.09083.

[BHL17] Baake M, Haynes A and Lenz D, Averaging almost periodic functions along ex-
ponential sequences. In Aperiodic Order. Vol. 2: Crystallography and Almost
Periodicity, M. Baake and U. Grimm (eds.), Cambridge University Press, Cam-
bridge (2017), pp. 343–362, arXiv:1704.08120.

[Bar17] Barreira L, Lyapunov Exponents, Birkhäuser, Basel (2017).

[BP07] Barreira L and Pesin Y, Nonuniform Hyperbolicity, Cambridge University Press,

Cambridge (2007).

[Ber66] Berger R, The undecidability of the domino problem, Memoirs Amer. Math. Soc.
66 AMS, Providence, RI (1966).

[Bes54] Besicovitch A S, Almost Periodic Functions, reprint, Dover, NY (1954).

[Boh93] Bohl P, Über die Darstellung von Functionen einer Variabeln durch
trigonometrische Reihen mit mehreren einer Variabeln proportionalen Argu-
menten, Master’s Thesis, Kaiserliche Universität zu Dorpat (1893).

[Boh47] Bohr H, Almost Periodic Functions, English edition, Chelsea, NY (1947).

19
[Dwo93] Dworkin S, Spectral theory and X-ray diffraction, J. Math. Phys. 34 (1993) 2965–
2967.

[EW11] Einsiedler M and Ward T, Ergodic Theory– with a View Towards Number Theory,
Springer, London (2011).

[Esc04] Esclangon E, Les Fonctions Quasi-périodiques, Dissertation, Université de Paris,

(1904).

[EW99] Everest G and Ward T, Heights of Polynomials and Entropy in Algebraic Dy-
namics, Springer, London (1999).

[FK60] Furstenberg H and Kesten H, Products of random matrices, Ann. Math. Stat. 31
(1960) 457–469.

[Fur63] Furstenberg H, Noncommuting random products, Trans. Amer. Math. Soc. 108
(1963) 377–428.

[FK83] Furstenberg H and Kifer Y, Random matrix products and measures on projective
spaces, Israel J. Math 46 (1983) 12–32.

[GL92] Godrèche C and Lançon F, A simple example of a non-Pisot tiling with five-fold
symmetry, J. Phys. I. France 2 (1992) 207–220.

[Hof95] Hof A, On diffraction by aperiodic structures, Commun. Math. Phys. 169 (1995)
25–43.

[Kin73] Kingman J F C, Subadditive ergodic theory, Ann. Prob. 1 (1973) 883–889.

[Man17a] Mañibo N, Lyapunov exponents for binary constant-length substitutions, J.

Math. Phys. 58 (2017) 113504 (9pp), arXiv:1706.00451.

[Man17b] Mañibo N, Spectral analysis of primitive inflation rules, Oberwolfach Rep. 14

(2017) 2830–2832.

[Man19] Mañibo N, Lyapunov Exponents in the Spectal Theory of Primitive Infla-

tion Systems, PhD Thesis, Bielefeld University (2019), online available at:
https://pub.uni-bielefeld.de/record/2935972.

[Ose68] Oseledets V I, A multiplicative ergodic theorem. Characteristic Ljapunov expo-

nents of dynamical systems (in Russian), Trans. Moscow Math. Soc. 19 (1968)
197–231.

[Pen74] Penrose R, The role of aesthetics in pure and applied mathematical research,
Bull. Inst. Math. Appl. 10 (1974) 266–271.

[Pol10] Pollicott M, Maximal Lyapunov exponents for random matrix products, Invent.
Math. 181 (2010) 209–226.

20
[SBGC84] Schechtman D, Blech I, Gratias D and Cahn J W, Metallic phase with long-
range orientational order and no translational symmetry, Phys. Rev. Lett. 53
(1984) 1951–1953.

[Via13] Viana M, Lectures on Lyapunov Exponents, Cambridge University Press, Cam-

bridge, (2013).

21