See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/291338344

Dynamical properties of the tent map

Article in Journal of the London Mathematical Society · August 2015

DOI: 10.1112/jlms/jdv071

CITATIONS READS

6 2,658

3 authors:

Klaus Scheicher Victor F. Sirvent

BOKU University Simón Bolívar University
51 PUBLICATIONS 559 CITATIONS 86 PUBLICATIONS 718 CITATIONS

SEE PROFILE SEE PROFILE

Paul Surer
BOKU University
18 PUBLICATIONS 108 CITATIONS

SEE PROFILE

All content following this page was uploaded by Klaus Scheicher on 21 January 2016.

The user has requested enhancement of the downloaded file.

 DYNAMICAL PROPERTIES OF THE TENT MAP

KLAUS SCHEICHER1 , VÍCTOR F. SIRVENT2 , AND PAUL SURER3

Abstract. Tent maps are continuous composites of two linear functions that act on
the unit interval. In the present paper we describe and analyse a connection between
dynamical systems induced by tent maps and the dynamics induced by a certain type of
beta-expansion. This relation, which is a weaker form of measure-theoretical conjugacy of
dynamical systems, allows us to transfer statements concerning the periodicity of orbits
but it turns out that the underlying symbolic dynamical systems are not connected via a
finite state transducer.

1. Introduction
Let α > 1 and define the tent map Tα ∶ [0, 1] → [0, 1] by
fL (x) for x ∈ [0, α1 ],
Tα (x) = {
fR (x) for x ∈ [ α1 , 1],
(1.1)

where
fL ∶ R Ð→ R, x z→ αx,
fR ∶ R Ð→ R, x z→ (1 − x).
α
α−1
The function Tα is continuous on (0, 1), its graph is sketched in Figure 1. Tent maps have
been intensively studied in [5, 6]. Apart from this, several results concerning the symmetric
case α = 2 can be found in [2, 3].
The intention of the present paper is to investigate the sets
Per(Tα ) ∶= {x ∈ [0, 1] ∶ ∃k2 > k1 ≥ 0 ∶ Tαk2 (x) = T k1 (x)},
Fin(Tα ) ∶= {x ∈ [0, 1] ∶ ∃k ≥ 0 ∶ Tαk (x) = 0},
especially, we want to know for which α > 1 we have
Per(Tα ) = Q(α) ∩ [0, 1], (P)
Fin(Tα ) = Z[α−1 ] ∩ [0, 1]. (F)
A lot of results concerning the periodicity property (P) have already been presented in
[5]. For example, it is known that (P) requires α to be an algebraic integer. A sufficient
condition for (P) is that α is a special Pisot number, which means that α as well as
β ∶= α(α − 1)−1 are Pisot numbers (a Pisot number is a real algebraic integer greater than 1
Date: August 19, 2015.
2010 Mathematics Subject Classification. 37B10, 11B85, 68R15.
1
2 K. SCHEICHER, V.F. SIRVENT, AND P. SURER

Figure 1. The Graph of the tent map Tα , here with α the golden mean. It
possesses the two fixed points 0 and ξ = 2α−1
α
.

such that the other Galois conjugates are contained in the complex unit circle). However,
a complete description of the algebraic integers that satisfy (P) is still outstanding. The
only general result here is the so-called U-number conjecture stated in [5].
Due to [10] there are only 11 special Pisot numbers. In the same paper it was shown that
there exist only one Salem number α such that β ∶= α(α − 1)−1 is a Pisot number. Recall
that a Salem number is a real algebraic integer greater than 1 such that the other Galois
conjugates have modulus smaller than or equal to 1 where equality holds at least once. As
a contribution concerning periodicity we show in the actual research that (P) holds for the
two associated tent maps Tα and Tβ . This result is a little surprising since in context with
beta-expansions the periodicity of Salem number systems is a famous problem (known as
Schmidt’s conjecture, see [9]) which remains unsolved.
In [6] three of the special Pisot numbers have been characterised in view of the finiteness
property (F). Here we will add several new results. More precisely, (F) will keep unverified
for only two special Pisot numbers where we strongly conjecture that it holds in these
cases.
Table 1 contains all special Pisot number as well as the Salem-Pisot pair, and summarises
all known results up to now concerning (P) and (F), the new ones and the less recent ones.

For confirming (F) we will use a connection of the tent map with a particular type of
beta-transformation. which is a weaker form of conjugation of dynamical systems. This
relation, which is a weaker form of conjugation of dynamical systems and does not seem to
be described in any other context yet, appears that interesting that we state it as principle
result of the present paper. In particular, define the set

S ∶= {α ∈ (1, 2] ∶ ∃P ∈ N ∶ ( ) = αP } .
α 2
α−1
 DYNAMICAL PROPERTIES OF THE TENT MAP 3

Number Approx. value Minimal polynomial (P) (F)

α−6 2.15372 . . . x4 − 3x3 + 3x2 − 3x + 1 ✓ —
α−5 4.0796 . . . x3 − 5x2 + 4x − 1 ✓ —
α−4 3.62966 . . . x − 5x3 + 6x2 − 4x + 1 ✓ —
4

α−3 3.1479 . . . x3 − 4x2 + 3x − 1 ✓ ?

α−2 2.61803 . . . x2 − 3x + 1 ✓ —
α−1 2.32472 . . . x3 − 3x2 + 2x − 1 ✓ ?
α0 2 x−2 ✓ ✓
α1 1.75488 . . . x3 − 2x2 + x − 1 ✓ ✓
α2 1.61803 . . . x −x−1
2 ✓ ✓
α3 1.46557 . . . x3 − x2 − 1 ✓ ✓
α4 1.38028 . . . x −x −1
4 3 ✓ —
α5 1.32472 . . . x −x−1
3 ✓ ✓
α6 1.86676 . . . x4 − 2x3 + x − 1 ✓ —

Table 1. The numbers α−5 to α5 are the 11 special Pisot numbers, the
numbers α−6 and α6 form the Salem-Pisot pair. For each i ∈ {−6, . . . , 6}
we have α−i = αi (αi − 1)−1 . All these numbers satisfy (P). The finiteness
property (F) only holds for some of the numbers and is still unverified for
α−1 and α−3 .

For α > 1 let β ∶= α(α − 1)−1 and note that the fixed point ξ of Tα is given by ξ = β(β + 1)−1 .
Define the interval I ∶= [ξ − 1, ξ) and the generalised fractional part function
φ ∶ R Ð→ I, x z→ x − ⌊x + (1 − ξ)⌋ .
Observe that the restriction of φ on the unit interval [0, 1] is almost bijective, i.e. bijective
up to a set of Lebesgue measure zero. Denote by τα the transformation
τα ∶ I Ð→ I, x z→ φ(αx). (1.2)
Observe that this transformation is actually a beta-transformation with respect to the
base α and the generalised fractional part function φ. Beta-expansions induced by such
transformations have been considered in [11] and fit into the framework studied in [4].
Theorem 1.1. Let α ∈ S. Then for each x ∈ [0, 1] there exist positive integers p = p(x), q =
q(x) such that φ ○ Tαp (x) = ταq ○ φ(x).
The relation can be illustrated by the following commutative diagram.
p(x)
[0, 1] [0, 1]
Tα

φ φ
q(x)
I = [ξ − 1, ξ) [ξ − 1, ξ)
τα
4 K. SCHEICHER, V.F. SIRVENT, AND P. SURER

Moreover, we show that there exists no upper bound for p and q, i.e., for each N ∈ N we
can find an x ∈ [0, 1] such that φ ○ Tαp (x) =/ ταq ○ φ(x) for all 1 ≤ p, q ≤ N .
Note that S is an infinite set with 1 as the only limit point. It contains the 6 special
Pisot numbers α0 to α5 . The following table shows the respecting values of P .
α0 α1 α2 α3 α4 α5
P 2 3 4 6 8 10
With the aid of Theorem 1.1 we are able to use algorithmic results presented in [11] in
order to verify (F) for these numbers.
The above mentioned connection between tent maps and beta-expansions motivates to
a closer analysis of the respective induced symbolic dynamics. The subshift induced by
tent maps is the full binary shift while the subshifts induced by the transformation τα are
of finite type for α ∈ S (this will follow from [4]). The relation from Theorem 1.1 allows us
to transfer results concerning the periodicity of orbits but it will turn out to be too weak
for the underlying subshifts to be connected via a finite state transducer.
The paper is organised as follows. In Section 2 we show that the symbolic shift space
induced by the tent map is the full binary shift. We also present the expansion of real
numbers with respect to the tent map. Section 3 deals with periodicity properties. We
recall known results and prove that the two tent maps associated with the Salem-Pisot
pair satisfy (P). Section 4 contains several relevant facts concerning the transformation τα
and the induced beta-expansions for α ∈ S. In Section 5 we prove Theorem 1.1. With this
result we are able to transfer finiteness properties from τα to Tα (and vice versa). This is
realised in Section 6. In Section 7 we are concerned with the underlying symbolic dynamics.
We characterise the generalised fractional part function φ in terms of the induced subshifts
and show that it is not given by a finite state transducer. Finally, Section 8 contains several
questions and problem that arose in context with our research.

2. Symbolic dynamics and representation of real numbers induced by the

tent map
In the present section we define the shift space associated with the tent map Tα and
show that it is the full binary shift. Furthermore, we recall and extend results from [5] in
order to represent real numbers with respect to Tα .
Let α > 1 and β = α−1α
. Define the map
L if x ∈ [0, α−1 ],
S ∶ [0, 1] Ð→ {L, R}, x z→ {
R if x ∈ (α−1 , 1].
Hence, Tα (x) = fS(x) (x) for all x ∈ [0, 1]. For x ∈ [0, 1] define the sequence d(x) ∶=
(S(Tαn (x)))n≥0 . Clearly, for x ∈ Fin(α) ∖ {0, 1} the sequence d(x) terminates with LRL∞ .
Furthermore, each arbitrary finite or infinite {R, L}-sequence appears as subword of d(x)
for some x ∈ [0, 1], with RRL∞ as the only exception. Of course, we also may include the
point α−1 in the interval corresponding to R but it would not have a significant impact on
the present paper. We leave the adaption of the results in that effect to the reader.
 DYNAMICAL PROPERTIES OF THE TENT MAP 5

The {R, L}-sequences of the elements of [0, 1] induce the one-sided subshift
W ∶= {d(x) ∶ x ∈ [0, 1]},
with the closure with respect to the product topology of the discrete topology. Recall that
a basis of this topology is given by the cylinder-sets. In particular, let W ∈ {L, R}∗ (the
set of finite words over {L, R}). Then the cylinder-set [W ] is defined by
[W ] ∶= {ω ∈ W ∶ ω is a proper prefix of W }.
Observe that cylinder-sets are clopen sets.
A suitable metric D ∶ W × W Ð→ R is given by
if ∀n ≥ 0 ∶ dn = d′n ,
D((dn )n≥0 , (d′n )n≥0 ) = {
0
exp(− min{n ∶ dn =/ d′n }) otherwise.
Lemma 2.1. The subshift W is the full shift {L, R}N .
Proof. It is clear that W ⊆ {L, R}N . On the other hand, if ω ∈ {L, R}N does not end with
RRL∞ then there exists an x ∈ [0, 1] such that d(x) = ω and we have ω ∈ W. Thus, we
have to concentrate on words of the shape ω = W RRL∞ with W ∈ {L, R}∗ . For each n ∈ N
define the sequence
ωn ∶= W RRLn R∞ .
Clearly, ωn ∈ W for each n ∈ N and we obviously have
lim ωn = ω,
n→∞

where the limit is with respect to the metric D. Hence, ω ∈ W. 

Note that d conjugates the tent map Tα on [0, 1] with the shift map σ on W, that is we
have σ ○ d(x) = d ○ Tα (x) for all x ∈ [0, 1].
Following [5], we can associate to each element of ω ∈ W an element v(ω) ∈ [0, 1] such
that v ○ d(x) = x for all x ∈ [0, 1]. In other words, v provides a way for representing real
numbers contained in the unit interval with respect to the tent map Tα .
Consider an element ω ∈ {L, R}N and count the L’s that occur between two consecutive
R’s. In particular, we denote by MR ∈ N ∪ {∞} the number of occurrences of R in ω and
by (lk )0≤k<MR the (finite or infinite) integer sequence that satisfies
if MR = ∞,
ω={ l
Ll0 RLl1 RLl2 R⋯
L 0 RLl1 R⋯LlMR −1 RL∞ otherwise.
Note that x ∈ Fin(α) if and only if MR < ∞.
Now, define v(ω) by
v(ω) ∶= ∑ (−β)−k α−(l0 +⋯+lk ) . (2.1)
0≤k<MR
Since α, β > 1 this series clearly converges. Furthermore, for each word W ∈ {L, R}∗ we
have v(W LRL∞ ) = v(W RRL∞ ). Observe that for MR = 0 the sequence (lk )0≤k<MR is the
empty sequence and (2.1) is the empty sum.
6 K. SCHEICHER, V.F. SIRVENT, AND P. SURER

Proposition 2.2. Let α > 1. Then the following properties hold.

(1) d is continuous on ([0, 1] ∖ Fin(Tα )) ∪ {0, 1}.
(2) d maps [0, 1] bijectively onto W ∖ O, where O denotes the inverse shift orbit of
RRL∞ , i.e.
O ∶= {ω ∈ {L, R}N ∶ ∃n ≥ 0 σ n (ω) = RRL∞ } . (2.2)
The inverse map is given by the evaluation map v.
(3) The evaluation map v maps W surjectively and continuously onto [0, 1].
Proof. (1) Since S is not continuous in α−1 we obviously have that d is not continuous in
any x whose Tα -orbit contains α−1 , which corresponds exactly with the set Fin(Tα )∖
{0, 1}. On the other hand, for each x ∈ ([0, 1] ∖ Fin(Tα )) ∪ {0, 1} we have that d
is continuous with respect to the discrete topology for each single digit and, hence,
continuous with respect to the product topology.
(2) By [5] we have that x = v ○d(x) is a unique representation of x. Hence, d is injective
and v the suitable inverse function. On the other hand, from Lemma 2.1 we know
that d([0, 1]) = W ∖ O. This shows the surjectivity.
(3) The surjectivity of v follows from the injectivity of v on W ∖ O. To show continuity
consider an arbitrary ω ∈ W and some arbitrary small z > 0. Let W we the initial
word of ω of length n = − ⌊logmin{α,β} (α−1 z)⌋. We claim that for each ω ′ ∈ [W ] we
have ∣v(ω) − v(ω ′ )∣ ≤ z, which immediately shows continuity.
We code ω and ω ′ as above, hence, MR , MR′ ∈ N ∪ {∞} denote the numbers of
occurrences of R and (lk )0≤k<MR , (lk′ )0≤k<MR′ are the respective integer sequences.
Furthermore, denote by WR and WL the number of occurrences of R and L, respec-
tively, in W . Note that WR ≤ MR as well as WR ≤ MR′ and we may suppose that at
least one of these inequalities is strict (otherwise we would have ω = ω ′ ). Now we
easily calculate
RRR RRR
RRR −k −(l0′ +⋯+lk′ ) RRR
∣v(ω) − v(ω )∣ =RR ∑ (−β) α − ∑ (−β) α RRR
RRR0≤k<MR
′ −k −(l0 +⋯+lk )

R 0≤k<MR′
RRR
RRR RRR
RRR −k −(l0′ +⋯+lk′ ) RRR
=RR ∑ (−β) α − ∑ (−β) α RRR
RRRWR ≤k<MR
−k −(l0 +⋯+lk )

R WR ≤k<MR
′ RRR
RRR RRR
−WR −WL RRR −k−WR −(l0′ +⋯+lk′ −WL ) RRR
≤β RRR ∑ (−β) − ∑ (−β) RRR.
−k−WR −(l0 +⋯+lk −WL )
α α α
RRRWR ≤k<MR WR ≤k<MR′ RRR
For each k with WR ≤ k < min{MR , MR′ } we clearly have
∣α−(l0 +⋯+lk −WL ) − α−(l0 +⋯+lk −WL ) ∣ ≤ 1.
′ ′

Thus,
∣v(ω) − v(ω ′ )∣ ≤ min{α, β}−n ∑ β −k ≤ zα−1 β(β − 1)−1 = zα−1 α = z.
k≥0


 DYNAMICAL PROPERTIES OF THE TENT MAP 7

In order to obtain v(ω) for a given ω ∈ W we also may count the R’s between the L’s. For
this purpose denote by ML ∈ N ∪ {∞} the number of occurrences of L in d(x). Analogously
to above we define the (finite or infinite) sequence (rk )0≤k<ML by
if ML = ∞,
ω={ r
Rr0 LRr1 LRr2 L⋯
R 0 LRr1 LRr2 L⋯RrML −1 LR∞ if ML < ∞.

j=0 (−β)
Observe that ξ = β(β + 1)−1 = ∑∞ ∑j=0 (−β)−j = ξ − (−β)−r ξ for all r ∈ N.
−j and r−1

This immediately yields

∑k=0L (−β)−(r0 +⋯+rk−1 ) α−k (−ξ + (−β)rk ξ) + α−M (−β)−(r0 +⋯+rm−1 ) ξ
M −1
if ML < ∞,
v(ω) = { ∞
∑k=0 (−β)−(r0 +⋯+rk−1 ) α−k (−ξ + (−β)rk ξ) if ML = ∞.
By a slight reorganisation of the addends and the observation that α−1 −1 = −β −1 we obtain
v(ω) = ξ + ξ ∑ (−β)−(1+r0 +⋯+rk ) α−k . (2.3)
0≤k<ML

We will need this representation in the proof of Theorem 3.4.

We see that ML = 0 is equivalent to x = ξ. For ML =/ 0 the integer r0 tells us something
about the position of x and ξ: the bigger n0 the smaller ∣x − ξ∣. Moreover, n0 even implies
x < ξ while n0 odd implies x > ξ. More precisely,
⎧
⎪[0, α−1 ] = [fR0 (0), fR−2 (0)]
⎪
⎪
⎪
if r0 = 0,
⎪
⎪
⎪(ξ − ξβ 0 , ξ − ξβ 0 ] = (fR (0), fR (0)] if r0 ≡ 0 (mod 2),
−r −r −2 −r 0 −r 0 −2
x∈⎨
[1 −1 + β −2 , 1] = (f −3 (0), f −1 (0)],
(2.4)
⎪
⎪
⎪ − β if r 0 = 1,
⎪ R R

⎩[ξ + ξβ
⎪
⎪
⎪ −r0 −2 , ξ + ξβ −r0 ) = [f −r0 −2 (0), f −r0 (0)), if r ≡ 1 (mod 2).
R R 0

3. Periodicity properties
We start the present section with a summary of what is already known about the peri-
odcity propery (P). At first we present a necessary condition.
Proposition 3.1 (cf. [5, Theorem 2.1]). Let α > 1 such that (P) holds. Then α and
α(α − 1)−1 are both algebraic integers and min{∣α̂∣, ∣α̂(α̂ − 1)−1 ∣} < 1 holds for all Galois
conjugates α̂ =/ α.
The next results states that special Pisot numbers satisfy (P).
Proposition 3.2 (cf. [5, Theorem 3.1]). Let α = αi for some i ∈ {−5, . . . , 5}. Then (P)
holds.
A more complete characterisation of numbers that satisfy (P) is still outstanding and
seems to be a hard problem. In [5] we can find the following conjecture which is based on
heuristic considerations.
Conjecture 3.3 (U-number conjecture). An algebraic integer α > 1 satisfies (P) if and only
if ∣α̂∣∣α̂ − 1∣ < 1 holds for all Galois conjugates α̂ =/ α.
1/α−1
8 K. SCHEICHER, V.F. SIRVENT, AND P. SURER

Observe that α−6 as well as α6 satisfy the condition of the U-number conjecture and,
hence, we already may expect them to satisfy (P). Indeed, the following new theorem
confirms this assertion.

Theorem 3.4. Property (P) is satisfied for α ∈ {α−6 , α6 }.

Proof. The technique of the proof comes from [9]. Since Per(Tα ) ⊆ Q(α) ∩ [0, 1] obviously
holds it suffices to show that for a given x = x0 ∈ [0, 1] ∩ Q(α) we have x ∈ Per(Tα ).
At first we treat the case α = α−6 , β = α(α − 1)−1 = α6 . Let ω ∶= d(x) and consider the
representation (2.1), i.e.

x0 = v(ω) = ∑ (−β)−k α−(l0 +⋯+lk ) .

0≤k<MR

If MR =/ ∞ we are obviously done. Hence, suppose that MR = ∞. For simplicity, define

mk ∶= − ∑ki=0 li for all k ≥ 0 and observe that that (mk )k≥0 is a monotonically decreasing
sequence with m0 ≤ 0. For j ≥ 1 define
∞ j
xj ∶= ∑ (−β)−k αmk+j −mj−1 = (−β)j α−mj−1 x0 − ∑ (−β)j−k αmk −mj−1 (3.1)
k=0 k=1

and note that xj = Tα j−1 (x) ∈ Q(α). Thus, we have ∣xj ∣ ≤ 1 for all j ≥ 0.
m +j

Now let α̂ =/ α be a Galois conjugate of α, β̂ ∶= α̂(α̂ − 1)−1 and denote for each j ≥ 0 the
respective algebraic conjugate of xj by x̂j ∈ Q(α̂). There are actually two possibilities. On
one hand α̂ as well as β̂ are real numbers with modulus strictly less than 1, on the other
hand both, α̂ and β̂, are complex numbers with ∣α̂∣ = 1 and ∣β̂∣ < 1. However, we obtain
from (3.1) for all j ≥ 1
j ∣β̂∣
∣x̂j ∣ ≤ ∣β̂∣ ∣α̂∣ ∣x̂0 ∣ + ∑ ∣β̂∣ ∣α̂∣ < ∣x̂0 ∣ +
j −mj−1 j−k mk −mj−1
.
k=1 1 − ∣β̂∣

This shows that the elements of the sequence (x̂j )j≥0 are bounded for all conjugates α̂ and,
thus, the number of possibilities is finite. Therefore, there exist two indices 0 > j1 > j2 such
that Tα j1 −1 (x) = xj1 = xj2 = Tα j2 −1 (x) and, hence, x ∈ Per(Tα ).
m +j1 m +j2

Now let α = α6 , β = α(α − 1)−1 = α−6 , and consider the representation (2.3), i.e.

x0 = v(ω) = ξ + ξ ∑ (−β)−(1+r0 +⋯+rk ) α−k .

0≤k<ML

Again we may assume that the sum has an infinite number of terms. Analogously to above
we define the decreasing integer sequence (mk )k≥0 by mk ∶= −1 − ∑ki=0 ri (hence, m0 ≤ −1).
For each j ≥ 1 define
∞
xj ∶= ξ + ξ ∑ (−β)mk+j −mj−1 +1 α−k (3.2)
k=0
 DYNAMICAL PROPERTIES OF THE TENT MAP 9

(x). We are interested in the sequence that is composed by

m −1+j
and note that xj = Tα j−1
yj ∶= xj ξ −1 − 1 for all j ∈ N. In particular, from (3.2) we obtain
∞
yj = ∑ (−β)mk+j −mj−1 +1 α−k
k=0

for all j ∈ N (where we formally set m−1 ∶= 0). Now we use the same argumentation as
above in order to show that there are two indices 0 > j1 > j2 such that yj1 = yj2 . This
(x) = xj1 = xj2 = Tα j2 −1 (x) and, hence, x ∈ Per(Tα ).
m −1+j1 m −1+j2
implies Tα j1 −1 

We already mentioned that in context with beta-expansions periodicity properties for

Salem number systems form a quite tricky problem. It is therefore remarkable that in our
setting, with α or β a Salem number, the proof runs quite uncomplicated. Note that there
does not exist any Salem number α such that β = α(α − 1)−1 is also a Salem number (see
[10, Theorem 2]).
Concerning the set S we can show that, apart from the six special Pisot numbers, none
of the elements induces a tent map that satisfies (P).
Lemma 3.5. Let α ∈ S. Then α = 2 or α is an algebraic unit. Furthermore, α is a Pisot
number if and only if β ∶= α(α − 1)−1 is a Pisot number. The periodicity property (P) is
satisfied if and only if α = αi for an i ∈ {0, . . . , 5}.
Proof. By the definition of S there exists an integer P ≥ 2 such that αP = β 2 . Hence, the
minimal polynomial of α must divide the polynomial xP −2 (x − 1)2 − 1. If P = 2 then we
have α = 2, otherwise this polynomial is monic with constant coefficient −1 and, therefore,
α is an algebraic unit.
Now consider an arbitrary Galois conjugate β̂ of β. Then there exists a corresponding
Galois conjugate α̂ of α such that α̂P = β̂ 2 and, hence, ∣β̂∣ = ∣α̂∣ . This shows that ∣β̂∣ < 1 if
P/2

and only if ∣α̂∣ < 1, ∣β̂∣ = 1 if and only if ∣α̂∣ = 1, and ∣β̂∣ > 1 if and only if ∣α̂∣ > 1. Especially,
α is a Pisot number if and only if β is a Pisot number (i.e., α is a special Pisot number).
The argumentation also shows that, except for the mentioned six special Pisot numbers,
no element of S satisfies the necessary conditions from Proposition 3.1. 

4. The dynamics induced by the transformation τα

In order to prove our main result, Theorem 1.1, in the next section we are going to
collect several important facts concerning the symbolic dynamics induced by τα (as defined
in (1.2)) for α ∈ S. A lot of results can be directly adopted from [4].
The transformation τα induces an expansion of real numbers that is closely related to
the well-known greedy-expansions introduced by Rényi in [8]. Let α ∈ S, β ∶= α(α − 1)−1 ,
ξ ∶= β(β + 1)−1 , and consider an x ∈ I = [ξ − 1, ξ). Successive application of τα yields an
infinite sequence δ(x) = (δk )k≥1 where δk = ⌊αταk−1 (x) + (1 − ξ)⌋. Since we are restricted to
α ∈ (1, 2] we have δ(x) ∈ {−1, 0, 1}N . To avoid confusion we write, from now on, 1̄ instead
of −1 in digit strings.
10 K. SCHEICHER, V.F. SIRVENT, AND P. SURER

Now, x can be represented as

x = ν(δ(x)) = ∑ δk α−k , (4.1)
k≥1

where, analogously to Section 2, ν denotes the evaluation function that assigns to each
sequence over {1̄, 0, 1} a real number. On the other hand, let (δk )k≥1 be an arbitrary
sequence with terms in {1̄, 0, 1}. We call the sequence admissible if there is some x ∈ I
such that δ(x) = (δk )k≥1 . A necessary and sufficient condition for admissibility is obviously
given by
∀m ≥ 1 ∶ ∑ δm+k α−k ∈ I.
k≥1

However, there is also a more advanced condition. Let τ̃α be the left-continuous trans-
formation corresponding to τα defined by
τ̃α ∶ (ξ − 1, ξ] → (ξ − 1, ξ] , x ↦ αx + ⌊−αx + ξ⌋ .
It is easy to see that for x ∈ R with αx − ξ ∈/ Z, we have τα (x) = τ̃α (x).
Proposition 4.1 (cf. [4, Theorem 2.5]). A sequence (δk )k≥1 ∈ {1̄, 0, 1}N is admissible if
and only if
δ(ξ − 1) ≤lex (δk )k≥m <lex δ̃(ξ)
holds for all m ≥ 1, where δ̃(y) ∶= (− ⌊−ατ̃αk−1 (y) + ξ⌋)k≥1 .
The set of all digit strings induces the subshift
Ω ∶= {δ(x) ∶ x ∈ I}
where the closure is with respect to the product topology of the discrete topology.
Proposition 4.2 (cf. [4, Proposition 2.14]). Let α ∈ S. A sequence (δk )k≥1 ∈ {−1, 0, 1}N is
contained in Ω if and only if
δ(ξ − 1) ≤lex (δk )k≥m ≤lex δ̃(ξ).
holds for all m ≥ 1 (with δ̃(ξ) as in Proposition 4.1). Moreover, the subshift Ω is sofic if
and only if δ(ξ − 1) as well as δ̃(ξ) are ultimately periodic.
Observe that, analogously to the tent map, δ conjugates the transformation τα on I with
the shift map σ on Ω, i.e. we have σ ○ δ(x) = δ ○ τα (x) for all x ∈ I.
In order to prove the soficness of Ω for α ∈ S we need the following lemma.
Lemma 4.3. Let 1 < α ≤ 2 and x ∈ I. Then we have
⎧
⎪αx + 1 if x < α−1 (ξ − 1),
⎪
⎪
⎪
τα (x) = ⎨αx − 1 if x ≥ α−1 ξ,
⎪
⎪
⎪
⎪
⎩αx otherwise.
 DYNAMICAL PROPERTIES OF THE TENT MAP 11

Proof. By the definition of τα we have τα (x) = αx−d with d ∈ Z such that d ≤ αx+1−ξ < d+1.
For x < α−1 (ξ − 1) we see that
(1 − α)(1 − ξ) ≤ αx + 1 − ξ < 0.
By observing that 1 − ξ = (α − 1)(2α − 1)−1 one easily √ verifies that the left hand expression
is bigger than or equal to −1 whenever 1 < α ≤ 2 + 2.
For x ≥ ξα−1 we have
1 ≤ αx + 1 − ξ < 1 + ξ(α − 1).
√
Similar as above we see that the right hand expression is smaller than 2 for 1 < α < 2 + 3.
For x ∈ [(ξ − 1)α−1 , ξα−1 ) we have
0 ≤ αx + 1 − ξ < 1
which concludes the proof. 
Proposition 4.4. Let α ∈ S. Then δ(ξ − 1) = (1̄10P −2 )∞ and δ̃(ξ) = (10P −1 )∞ . Especially,
the subshift Ω is sofic.
Proof. We start with the calculation of δ(ξ−1). By Lemma 4.3 we have τα (ξ−1) = α(ξ−1)+1
and δ(ξ − 1) starts with 1̄.
For the following calculations we need the inequality α ≤ 2 as well as the identities
α−1 + β −1 = 1 and β(1 − ξ) = ξ.
τα (ξ − 1) = α(ξ − 1) + 1 = (β + 1 − α)(1 − ξ) ≥ (β − 1)(1 − ξ) = βα−1 (1 − ξ) = α−1 ξ.
Thus, by Lemma 4.3, the second term of δ(ξ − 1) equals 1. Furthermore,
τα2 (ξ − 1) = (αβ + α − α2 − β − 1)(1 − ξ) = (α − 1)2 (ξ − 1) = α2 β −2 (ξ − 1) = α2−P (ξ − 1).
For P = 2 (and, hence, α = 2), we are done. Otherwise we obviously have τα2+k (ξ − 1) =
α2−P +k (ξ − 1) for all k ∈ {1, . . . , P − 2} and the (k + 2)nd term of δ(ξ − 1) is 0. Since
ταP (ξ − 1) = ξ − 1 we see that δ(ξ − 1) = (1̄10P −2 )∞ is purely periodic.
For the calculation of δ̃(ξ) we observe
τ̃α (ξ) = αξ + ⌊−αξ + ξ⌋ = αξ − 1
since −1 < −αβ −1 ξ = (1 − α)ξ < 0. This yields 1 as first term. From this we easily calculate
αβ − β − 1 α − 1
τ̃α (ξ) = αξ − 1 =
α β ξ
= = = P −1 = P −1 .
β+1 β + 1 β(β + 1) α (β + 1) α
As α−j ξ ≤ α−1 ξ for non-negative j we immediately obtain that τ̃αP (ξ) = ξ and δ(ξ) =
(10P −1 )∞ .
The soficness follows from Proposition 4.2. 
As Ω is sofic it can be described by a finite graph (see Figure 2). Actually, we are even
able to show a stronger result.
Proposition 4.5. Let α ∈ S. Then the subshift Ω is of finite type.
12 K. SCHEICHER, V.F. SIRVENT, AND P. SURER

Proof. The set

⋃ {V ∈ {1̄, 0, 1}j ∶ (V <lex 1̄10P −2 ) ∨ (V >lex 10P −1 )}

P

j=1

is a finite set of forbidden words. 

1̄10P −2
1̄10P −2 1̄10P −2
1̄10P −2
1 0 0
0 1 2 3 P −1 P 1̄10P −2

Figure 2. The subshift Ω can be represented by a finite graph with P

vertices. Each vertex has one edge to vertex P labelled by 1̄10P −2 . Addi-
tionally, vertex 1 has one self-loop labelled by 0 and one edge to vertex 2
labelled by 1, vertex P has one edge to vertex 1 labelled by 0, and, for each
j ∈ {2, 3, . . . , P − 1}, the vertex j has one edge to j + 1 labelled by 0.

Analogously to Proposition 2.2 we summarise several important properties concerning

δ and ν. We omit the proofs since they run analogously to Proposition 2.2. Additionally,
one can draw on well-known results concerning subshifts induced by the greedy-expansion.
Proposition 4.6. Let α ∈ S. Then the following items hold.
(1) δ is continuous on I ∖ {x ∈ I ∶ ∃n ≥ 1 ∶ ταn (x) = ξ − 1}.
(2) δ maps I bijectively onto Ω∖U, where U denotes the inverse shift orbit of (10P −1 )∞ ,
i.e.
U ∶= {ω ∈ Ω ∶ ∃n ≥ 0 ∶ σ n (ω) = (10P −1 )∞ } . (4.2)
The inverse map is given by the evaluation map ν.
(3) The evaluation map ν maps Ω surjectively and continuously onto I = [ξ − 1, ξ].

5. Proof of the main result

In the present section we prove Theorem 1.1 stated in the Introduction. The proof
is quite technical and requires some preliminary considerations. Recall that, for a given
α > 1, we defined β ∶= α(α − 1), ξ ∶= β(β + 1)−1 and I = [ξ − 1, ξ). We start with a suitable
subdivision of the intervals [0, 1] and I, respectively. Define
J−1 ∶= [0, α−1 ξ), I−1 ∶= φ(J−1 ) = [0, α−1 ξ),
′
J−1 ∶= [α−1 ξ, α−1 ], ′
I−1 ∶= φ(J−1
′
) = [α−1 ξ, α−1 ],
J0 ∶= (α−1 , ξ), I0 ∶= φ(J0 ) = (α−1 , ξ),
J0′ ∶= [ξ, 1 − α−1 β −1 ), I0′ ∶= φ(J0′ ) = [ξ − 1, −α−1 β −1 ),
 DYNAMICAL PROPERTIES OF THE TENT MAP 13

and for each n ≥ 1

Jn ∶= [1 − α−n β −1 , 1 − α−n (1 − ξ)), In ∶= φ(Jn ) = [−α−n β −1 , −α−n (1 − ξ)),
Jn′ ∶= [1 − α−n (1 − ξ), 1 − α−n−1 β −1 ), In′ ∶= φ(Jn ) = [−α−n (1 − ξ), −α−n−1 β −1 ).
One easily verifies that each interval has non-empty interior and
[0, 1] = ⋃ (Jn ∪ Jn′ ) ∪ {1}, I = ⋃ (In ∪ In′ ), (5.1)
n≥−1 n≥−1

where the unions are disjoint. Furthermore, S(x) = L for x ∈ J−1 ∪ J−1 ′
, otherwise we have
S(x) = R. For τα we already showed some rules in Lemma 4.3. Figure 3 sketches the
shape and the position of the intervals as well as the first digit of the respecting sequence
of digits.

Figure 3. The position and shape of the intervals In ,, In′ , Jn , and Jn′ for
n ≥ −1. The digit sequence d(x) (δ(x), respectively) starts with the indicated
digits whenever x is contained in the respecting intervals.

We are now in the position to show Theorem 1.1. In fact, in the following theorems
we will prove a much more detailed result, namely, the validity of Table 2 that shows the
values of p and q in terms of the intervals containing x. The Table also shows the first p
and q, respectively, digits of the corresponding digit strings.

Case x∈ φ(x) ∈ (p, q) (dk )1≤k≤p (δk )1≤k≤q

1 J−1 I−1 (1, 1) L 0
2 ′
J−1 ′
I−1 (1, 1) L 1
3 J0 I0 (2, P ) RR 10P −1
4 Jn (n ≥ 1) In (2 + n, P + n) RLn R 0n 1̄10P −2
5 Jn′ (n ≥ 0) In′ (2 + n, P + n) RLn R 0n−1 1̄10P −1
6 {1} {0} (1, 1) R 0

Table 2. For x in the specified set we have φ ○ Tαp (x) = ταq ○ φ(x). The fifth
column shows the first p terms of (dk )k≥1 ∶= d(x), the rightmost one the first
q terms of (δk )k≥1 ∶= δ(φ(x)).

The first theorem immediately implies Theorem 1.1.

14 K. SCHEICHER, V.F. SIRVENT, AND P. SURER

Theorem 5.1. Let α ∈ S and x ∈ [0, 1]. Then φ ○ Tαp (x) = ταq ○ φ(x) where the pair (p, q)
can be obtained from Table 2.

Proof. We distinguish the cases with respect to Table 2.

Cases 1 and 2: x ∈ J−1 ∪ J−1
′
. We easily see that φ ○ Tα (x) = φ(αx) = τα (x) = τα ○ φ(x) in
this case. Hence, the assertion holds for p = q = 1.
For the other cases we do the calculations in a very detailed way. Especially, the respective
calculations will also contain the exact images of the respecting intervals. In this way we
can use Figure 3 to determine the action of Tα and τα , respectively.
Case 3: x ∈ J0 . Without difficulties we calculate
Tα (x) = β(1 − x) ∈ (ξ, 1) ,
Tα2 (x) = β(1 − Tα (x)) ∈ (0, ξ) .

Hence,
φ ○ Tα2 (x) = β(1 − β(1 − x)) = β 2 x − β 2 + β.
On the other hand, observe that φ(x) = x ∈ I0 . By Lemma 4.3 we have τα (x) =
αx − 1. The corresponding interval can be easily calculated by observing the proof
of Proposition 4.4. Thus,
τα (x) = αx − 1 ∈ (0, ξα1−P ) ,
τα2 (x) = α(αx − 1) ∈ (0, ξα2−P ) ,
⋮ ⋮ ⋮
τα (x) =
P −1 α −2 (αx − 1)
P ∈ (0, ξα−1 ) ,
ταP (x) = αP −1 (αx − 1) ∈ (0, ξ) ,

Therefore,

ταP ○ φ(x) = αP x − αP −1 = β 2 x − β 2 α−1 = β 2 x − β 2 + β,

where we used α−1 = 1 − β −1 for the last step. We see that φ ○ T p (x) = τ q ○ φ(x)
holds for (p, q) = (2, P ).
Case 4: x ∈ Jn for n ≥ 1. By keeping the identity β(1 − ξ) = ξ in mind we conclude that

Tα (x) = β(1 − x) ∈ (ξα−n , α−n ] ,

Tα2 (x) = αTα (x) ∈ (ξα−n+1 , α−n+1 ] ,
⋮ ⋮ ⋮
Tαn (x) = αn−1 Tα (x) ∈ (ξα−1 , α−1 ] ,
Tαn+1 (x) = αn Tα (x) ∈ (ξ, 1] ,
Tαn+2 (x) = β(1 − Tαn+1 (x)) ∈ [0, ξ) .

Hence,
φ ○ Tαn+2 (x) = β(1 − αn β(1 − x)) = β 2 αn x − β 2 αn + β.
 DYNAMICAL PROPERTIES OF THE TENT MAP 15

Now we consider the action of τα on φ(x) = x − 1 ∈ In . By Lemma 4.3 we

immediately obtain
τα (x − 1) = α(x − 1) ∈ [−α1−n β −1 , α1−n (ξ − 1)) = In−1 ,
⋮ ⋮ ⋮
τα (x − 1) = α (x − 1)
n−1 n−1 ∈ [−α−1 β −1 , α−1 (ξ − 1)) = I1 ,
τα (x − 1) = α (x − 1) + 1 ∈ [α−1 , ξ).
n n

The interval [α−1 , ξ) coincides, up to the boundary, with I0 . One easily verifies that
we can apply Case 3 without any problems. This yields
ταn+P (x − 1) = αP −1 (αταn (x − 1) − 1) ∈ [0, ξ),
and finally
ταn+P ○ φ(x) =β 2 ταn (x − 1) − β 2 αn + β
=β 2 αn x + β 2 − β 2 αn − β 2 + β = β 2 αn x − β 2 αn + β.

Therefore, the pair (p, q) = (n + 2, n + P ) satisfies the assertion in this case.

Case 5: x ∈ Jn′ for n ≥ 0. We observe that β(1 − ξ) = ξ and calculate
Tα (x) = β(1 − x) ∈ (α−n−1 , ξα−n ] ,
Tα2 (x) = αTα (x) ∈ (α−n , ξα1−n ] ,
⋮ ⋮ ⋮
Tαn (x) = αn−1 Tα (x) ∈ (α−2 , ξα−1 ] ,
Tαn+1 (x) = αn Tα (x) ∈ (α−1 , ξ] ,
Tαn+2 (x) = β(1 − Tαn+1 (x)) ∈ [ξ, 1) .
Hence,
φ ○ Tαn+2 (x) = β(1 − αn β(1 − x)) − 1 = β 2 αn x − β 2 αn + β − 1.
Quite analogously to the previous case we apply τα on φ(x) = x − 1 ∈ In′ . By
observing Lemma 4.3 we obtain
τα (x − 1) = α(x − 1) ∈ [α1−n (ξ − 1), −α−n β −1 ) = In−1
′
,
⋮ ⋮ ⋮
ταn (x − 1) = αn (x − 1) ∈ [ξ − 1, −α−1 β −1 ) = I0′ .
Considering the calculation of δ(ξ − 1) in Propositions 4.4 gives
ταn+1 (x − 1) = αταn (x − 1) + 1 ∈ [α(ξ − 1) + 1, α−1 ).
ταn+2 (x − 1) = αταn+1 (x − 1) − 1 ∈ [α−P +2 (ξ − 1), 0),
ταn+3 (x − 1) = αταn+2 (x − 1) ∈ [α−P +3 (ξ − 1), 0),
⋮ ⋮ ⋮
ταn+P −1 (x − 1) = αP −3 ταn+2 (x − 1) ∈ [α−1 (ξ − 1), 0),
ταn+P (x − 1) = αP −2 ταn+2 (x − 1) ∈ [ξ − 1, 0).
16 K. SCHEICHER, V.F. SIRVENT, AND P. SURER

Joining these results appropriately yields

ταn+P ○ φ(x) =αP −2 (α(αn+1 (x − 1) + 1) − 1) = αn+P x − αn+P + αP −1 − αP −2
=β 2 αn x − β 2 αn + αP −1 (1 − α−1 ) = β 2 αn x − β 2 αn + βα−1
=β 2 αn x − β 2 αn + β − 1.
We thus have shown that that φ ○ Tαp (x) = ταq ○ φ(x) holds for (p, q) = (n + 2, n + P ).
Case 6: x = 1. Here (p, q) = (1, 1) obviously serves.

The detailed realisation of the proof of Theorem 5.1 allows us to obtain further results
without too much additional effort. The next theorem shows that the values for p and q
are almost minimal in our context.
Theorem 5.2. Let α ∈ S, J an element of the collection {Jn ∶ n ≥ −1} ∪ {Jn′ ∶ n ≥ −1},
and (p, q) the corresponding pair with respect to Table 2. Then for all but finitely many
elements x ∈ J we have φ ○ Tαp (x) =/ ταq ○ φ(x) for all pairs (p′ , q ′ ) =/ (p, q) with 1 ≤ p′ ≤ p
′ ′

and 1 ≤ q ′ ≤ q.
Proof. If (p, q) = (1, 1) (Cases 1 and 2) the assertion is a triviality. Otherwise, consider the
(finite) sequences of intervals (φ ○ Tαp (J))1≤p′ ≤p and (ταq ○ φ(J))1≤q′ ≤q . Observe that we can
′ ′

easily obtain these sequences from the proof of Theorem 5.1 for the different cases.
At first note that φ ○ Tαp (J) = ταq ○ φ(J) holds if and only if (p′ , q ′ ) = (p, q). But
′ ′

we cannot exclude pointwise coincidences in these cases. However, since for a given pair
(p, q) =/ (p′ , q ′ ) with p′ ∈ {1, . . . , p} and q ′ ∈ {1, . . . , q} the maps φ ○ Tαp as well as ταq ○ φ
′ ′

are affine transformations of J, we see that φ ○ Tαp (x) = ταq ○ φ(x) holds for at most one
′ ′

element x ∈ J. 
As immediate consequence of the theorem we obtain that, in general, a stronger form of
Theorem 1.1 does not hold.
Corollary 5.3. Let α ∈ S. Then, for each N ∈ N we can find a set J ⊂ [0, 1] of positive
measure such that for all x ∈ J we have φ ○ Tαp (x) =/ ταq ○ φ(x) for all 0 < p, q ≤ N .
Proof. For an arbitrary integer n > N let J ∶= Jn ∪ Jn′ , i.e. J is an interval. By Theorem 5.2
we have φ ○ Tαp (x) =/ ταq ○ φ(x) for all but finitely many x ∈ J whenever 1 ≤ p < n + 2 and
1 ≤ q < n + P. 
The last result we will need in Section 7, where we analyse the action of φ in terms of
the subshifts W and Ω.
Theorem 5.4. Let α ∈ S, J an element of the collection {Jn ∶ n ≥ −1} ∪ {Jn′ ∶ n ≥ −1},
and (p, q) the corresponding pair with respect to Table 2. Then, for each x ∈ J the first p
terms of d(x) are given in the fifth column and the first q digits of δ ○ φ(x) are given in the
rightmost column of Table 2.
Proof. This can be seen immediately from the proof of Theorem 5.1. 
 DYNAMICAL PROPERTIES OF THE TENT MAP 17

6. Finiteness properties
In this section we deal with the finiteness property (F) of the tent map Tα for α contained
in Table 1. The first result in this context is already known.
Proposition 6.1 (cf. [6, Theorem 3.1]). For the symmetric tent map we have
Fin(T2 ) = {2−n m ∶ n, m ∈ N, 2n ≤ m},
especially, T2 satisfies (F).
In general, disproving (F) is much easier than verifying it since we only need counterex-
amples, i.e. elements of Z[α−1 ] ∩ (0, 1] that are purely periodic. Note that all numbers in
Table 1, except α0 , are algebraic units. Hence we have Z[α−1 ] = Z[α] in these cases.
Proposition 6.2. The tent map Tα does not satisfy (F) for α = αi with i ∈
{−6, −5, −4, −2, 4, 6}.
Proof. It suffices to show that for each of the six cases there exist (non-trivial) purely
periodic elements. In particular, one readily verifies the periodicity of the following points.
α = α−6 : For x = α3 − 2α2 + α − 2 we have Tα3 (x) = x.
α = α−5 : Here ξ is contained in Z[α] since ξ = 2α2 − 9α + 4, and Tα (ξ) = ξ.
α = α−4 : Again, ξ is contained in Z[α] since ξ = 4α3 − 18α2 + 15α − 8.
α = α−2 : As it was already observed in [6], we have ξ ∈ Z[α] with ξ = α − 2.
α = α4 : For x = α2 − 1 we have Tα4 (x) = x.
α = α6 : For x = 2α3 − 3α2 − 2α + 2 we have Tα4 (x) = x.

In order to confirm (F) for the elements of S we will exploit the relation elaborated in
Theorem 1.1. Analogously to the tent map Tα we define finiteness and, for completeness,
periodicity conditions for τα . Let
Per(τα ) ∶= {x ∈ I ∶ ∃k2 > k1 ≥ 0 ∶ ταk2 (x) = ταk1 (x)},
Fin(τα ) ∶= {x ∈ I ∶ ∃k > 0 ∶ ταk (x) = 0},
and consider the conditions
Per(τα ) = Q(α) ∩ I, (P-τ )
Fin(τα ) = Z[α−1 ] ∩ I. (F-τ )
Theorem 6.3. Let α ∈ S. Then the following items hold.
(i) x ∈ Per(Tα ) if and only if φ(x) ∈ Per(τα );
(ii) x ∈ Fin(Tα ) if and only if φ(x) ∈ Fin(τα ).
Proof. Let x ∈ [0, 1]. By successive application of Theorem 1.1 we can find strictly increas-
ing integer sequences (pk )k≥0 and (qk )k≥0 such that φ ○ Tαpk (x) = ταqk ○ φ(x) holds for all
k ≥ 0.
Suppose first that x ∈ Per(Tα ). Hence, the sequence (Tαn (x))n≥0 obtains only finitely
many different values. Therefore, the subsequence (Tαpk (x))k≥0 obtains only finitely many
18 K. SCHEICHER, V.F. SIRVENT, AND P. SURER

values. We thus can find integers k1 , k2 with 0 ≤ pk1 < pk2 such that Tα k1 (x) = Tα k2 (x).
p p

Hence, ταk1 ○ φ(x) = ταk2 ○ φ(x) and φ(x) ∈ Per(τα ).

q q

Now consider the case x ∈ Fin(Tα ). This implies the existence of an n0 such that
Tαn (x) = 0 for all n ≥ n0 . Thus, we can find an index k0 with n0 ≤ pk0 such that Tα k0 (x) = 0.
p

Hence, τα ○ φ(x) = φ ○ Tα (x) = φ(0) = 0, which shows that φ(x) ∈ Fin(τα ).

qk0 pk0

The contrary statement can be shown analogously with the additional observation that
φ is injective except for x ∈ {0, 1}. But this shall not bother since 0, 1 ∈ Fin(Tα ) and
φ(0) = φ(1) = 0 ∈ Fin(τα ). 
With this theorem the equivalence of the properties (P) and (P-τ ) ((F) and (F-τ ),
respectively) immediately follows for α ∈ S.
Corollary 6.4. Let α ∈ S. Then Tα satisfies (P) if and only if τα satisfies (P-τ ) and Tα
satisfies (F) if and only if τα satisfies (F-τ ).
Proof. This follows immediately by Theorem 6.3 and the observation that the restriction of
φ on the set Q(α) (Z[α−1 ], respectively) is a measure-theoretically bijection from Q(α) ∩
[0, 1] onto Q(α) ∩ I (Z[α−1 ] ∩ [0, 1] onto Z[α−1 ] ∩ I, respectively), where bijection fails only
for points that are contained in Fin(Tα ) and Fin(τα ), respectively. 
Hence, in order to prove (F) it suffices to show (F-τ ). The tools for this are so-called
shift radix systems. This notion of dynamical systems has been introduced in [1]. We need
a slightly generalised form that has been considered in [11]. We quickly summarise the
most important facts.
Definition 6.5 (cf. [1, 11]). Let s = (s0 , . . . , sn−1 ) ∈ Rn , ε ∈ [0, 1), and define the map
θ ∶ Zn Ð→ Zn by
θ ∶ (x0 , . . . , xn−1 ) z→ (x1 , . . . , xn−1 , − ⌊x0 s0 + ⋯ + xn−1 sn−1 + ε⌋).
We call the dynamical system (Zn , θ) a shift radix system if for all x ∈ Zd there exists an
integer n > 0 such that θ n (x) = 0.
For certain parameters the notion of shift radix system is intimately related with our
finiteness property (F-τ ).
Theorem 6.6 (cf. [11, Theorem 3.4]). Let α be an algebraic integer of degree n + 1 and
suppose that
(xn + sn−1 xn−1 + ⋯ + s0 )(x − α) ∈ Z[x]
is its minimal polynomial. Then, for s = (s0 , . . . , sn−1 ) ∈ Rn and ε = 1−ζ = (α−1)(2α−1)−1 ,
the pair (Zn , θ) is a shift radix system if and only if τα satisfies (F-τ ).
In order to verify whether a pair of parameters (s, ε) induces a shift radix system we
need some further definitions.
Definition 6.7 (cf. [1, 11]). Let s = (s0 , s1 , . . . , sn−1 ) ∈ Rn . A set V ⊂ Zn that
● contains the positive and negative Cartesian unit vectors and
 DYNAMICAL PROPERTIES OF THE TENT MAP 19

● for each (x0 , x1 , . . . , xn−1 ) ∈ V the vectors (x1 , . . . , xn−1 , − ⌊x0 s0 + ⋯ + xn−1 sn−1 ⌋) as
well as (x1 , . . . , xn−1 , − ⌈x0 s0 + ⋯ + xn−1 sn−1 ⌉) are contained in V , too,
is called a set of witnesses of s.
Note that for (s0 , s1 , . . . , sn−1 ) we can always find a finite set of witnesses provided that
the roots of the polynomial xn + sn−1 xn−1 + ⋯ + s0 are contained in the complex unit circle.
With a (finite) set of witnesses of a given s ∈ Rn it is very easy to verify whether we have
a shift radix system.
Theorem 6.8 (cf. [11, Theorem 2.7]). Let s ∈ Rn , V a set of witnesses of s, and ε ∈ [0, 1).
The pair (Zn , θ) is a shift radix system if and only if for all x ∈ V there exists an n ∈ N
such that θ n (x) = 0.
With this result we are finally in the position to prove (F) for several elements of S.
Theorem 6.9. Let α = αi for i ∈ {1, 2, 3, 5}. Then (F) holds. Furthermore, (F) does not
hold for any α ∈ S ∖ {α0 , α1 , α2 , α3 , α5 }.
Proof. The case α = α2 has already be shown in [6, Corollary 3.2b]. Note that the used
method cannot be generalised that easy.
For α = αi with i ∈ {1, 3, 5} set ε = 1 − ζ = (α − 1)(2α − 1)−1 and note that the algebraic
degree of α equals 3. By Corollary 6.4 and Theorem 6.6 it suffices to show that for a certain
parameter s the pair (Z2 , θ) is a shift radix system. We will do this by using Theorem 6.8.
In particular,
α = α1 : Following Theorem 6.6 we are interested in s = (α−1 , α −2). One easily verifies that
V ∶= {(0, 0), ±(1, 0), ±(0, 1), ±(1, 1), ±(1, −1)}
is a set of witnesses of s and for each x ∈ V there exists a positive integer n such
that θ n (x) = 0. Hence, by Theorem 6.8, (Z2 , θ) is a shift radix system, τα satisfies
(F-τ ) by Theorem 6.6, and Tα satisfies (F) by Corollary 6.4.
α = α3 : We set s = (α−1 , α − 1) and observe that a set of witnesses is given by
V ∶= {(0, 0, ), ±(1, 0), ±(0, 1), ±(1, 1), ±(1, −1), ±(2, 0), ±(0, 2), ±(2, −1), ±(−1, 2)}.
It is easy to check that for each x ∈ V the θ-orbit is eventually 0. This again implies
that Tα satisfies (F).
α = α5 : Here we have s = (α−1 , α). Then
V ∶= {(0, 0, ), ±(1, 0), ±(0, 1), ±(1, 1), ±(1, −1), ±(2, 0), ±(0, 2), ±(2, −1),
± (−1, 2), ±(2, 1), ±(2, −2), ±(3, 0), ±(0, 3), ±(3, −1), ±(1, −3), ±(3, −2),
± (2, −3), ±(3, −3), ±(4, −2), ±(4, −3), ±(3, −4), ±(4, −4)}
is set of witnesses for s. Successive application of θ on each element of V yields 0,
hence, (F) holds for Tα .
We already showed that (F) is satisfied for α = α0 while for α = α4 it does not hold.
In Lemma 3.5 we observed that for the other elements of S (P) does not hold since the
necessary conditions stated in Proposition 3.1 are not satisfied. From the respective original
20 K. SCHEICHER, V.F. SIRVENT, AND P. SURER

proof of [5, Theorem 2.1] we easily see that, actually, (F) implies (P) and, hence, (F) cannot
hold for any α ∈ S different from a special Pisot number . 

7. Relation between the dynamical systems W and Ω

In Theorem 1.1 we showed for α ∈ S the interaction between the tent map Tα , the
transformation τα and the generalised fractional part φ. This result induces in a natural
way a function Φ ∶ W Ð→ Ω such that δ ○ φ(x) = Φ ○ d(x) holds for all x ∈ [0, 1]. In the
present section we want to concentrate on this function Φ.
By Proposition 2.2 and Proposition 4.6 we see that Φ(ω) = δ ○ φ ○ v(ω) must hold for
all ω ∈ W ∖ O (with O as defined in (2.2)). We define the function Φ to be the continuous
extension on W, hence, by Proposition 2.2 and Proposition 4.6 we define Φ on W by the
the well defined composition
Φ ∶ W Ð→ Ω, ω z→ δ ○ φ ○ v(ω).
The following diagram explains the situation graphically.
W Φ Ω
d v δ ν
[0, 1]
φ
I
By Theorem 1.1 we can find for each x ∈ [0, 1] positive integers p, q such that φ ○ Tαp (x) =
ταq ○ φ(x). Thus, by the definition of Φ, we immediately obtain
Φ ○ σ p ○ d(x) = σ q ○ δ ○ φ(x).
This shows that we can characterise Φ by substituting blocks of symbols. However, we will
see that Φ is not given by a block-code in the sense of [7] and, more general, cannot be
described by a finite state transducer.
Table 2 in Section 5 already gives us an idea of the substitution rules. The sets involved
in the decomposition of [0, 1] (see (5.1)) play a major role. We are interested in their
images under d. Denote for each integer n ≥ −1 the word Wn ∈ {R, L}∗ defined by

if n = −1,
Wn ∶= { n
L
RL R otherwise.
We immediately obtain the following lemma.
Lemma 7.1. For all n ≥ −1 we have d(Jn ∪ Jn′ ) ⊂ [Wn ]. Furthermore, W = {RL∞ } ∪
⋃n≥−1 [Wn ] is a disjoint union.
Proof. The first part follows directly from Table 2. Note that the inclusion is strict since all
cylinder-sets [Wn ] include sequences that are contained in O, i.e. terminate with RRL∞ ,
which do not occur as d-image for any x ∈ [0, 1]. The second statement is trivial. 
 DYNAMICAL PROPERTIES OF THE TENT MAP 21

The main difficulty is that the d-images of Jn and Jn′ are not cylinder-sets for any n ≥ −1.
Define for each integer n ≥ −1
Jn ∶= ⋃ [Wn R2k L], Jn′ = ⋃ [Wn R2k+1 L].
k≥0 k≥0
For simplicity we set
J ∶= {Jn ∶ n ≥ −1} ∪ {Jn′ ∶ n ≥ −1}.
Lemma 7.2. The elements of the collection J are open sets that are pairwise disjoint. For
each integer n ≥ −1 we have [Wn ] ∖ (Jn ∪ Jn′ ) = {Wn R∞ }.
Proof. Each element of J is the infinite union of cylinder-sets and, thus, open. The other
statements are trivial. 
Lemma 7.3. For each integer n ≥ −1 the function d bijectively maps Jn onto Jn ∖ O and
Jn′ onto Jn′ ∖ O.
Proof. From Proposition 2.2 we know that d maps [0, 1] bijectively onto {L, R}N ∖ O. By
Lemma 7.1 we have
W ∖ O = {RL∞ } ∪ ⋃ ([Wn ] ∖ O)
n≥−1
is a disjoint union and
[0, 1) = ⋃ (Jn ∪ Jn′ )
n≥−1
is also a disjoint union. From these observations we conclude, by using Table 2, that d
maps Jn ∪ Jn′ bijectively onto [Wn ] ∖ O for all integers n ≥ −1 (and, of course, d(1) = RL∞ ).
Hence, it suffices to show that for all n ≥ −1 we have d(Jn ) ⊆ Jn and d(Jn′ ) ⊆ Jn′ .
Let n ≥ −1. We claim that that Jn′ = Jn′ ∪ {Wn R∞ }. At first note that Wn R∞ ∈ Jn′ and,
therefore, Jn′ ⊃ Jn′ ∪{Wn R∞ }. To show the opposite direction, observe that, by Lemma 7.2,
we have Jn ∪ Jn′ ∪ {Wn R∞ } = [Wn ] and, hence, since Jn ∩ Jn′ = ∅, we have
Jn′ ∪ {Wn R∞ } = [Wn ] ∖ Jn .
The right hand side is a closed set since it is the set difference of a closed set and an
open set. Hence, it coincides with its closure. From this we immediately obtain the other
inclusion
Jn′ ⊂ Jn′ ∪ {Wn R∞ } = Jn′ ∪ {Wn R∞ }.
We want to remark that, analogously, Jn = Jn ∪ {Wn R∞ } holds.
Now observe that from (2.4) it follows that x > ξ implies that d(x) starts with an odd
number of R’s before the first L occurs, while d(ξ) = R∞ . From the proof of Theorem 5.1
we known that for x ∈ Jn′ we have Tαp (x) ∈ [ξ, 1], hence, d(Jn′ ) ⊆ Jn′ . On the other hand,
d(x) starts with an even number of R’s for x < ξ and Tαp (Jn ) ⊂ [0, ξ), thus, d(Jn ) ⊆ Jn . 
Now we can characterise our function Φ by a block substitution (see Table 3), where the
respective substitution rule depends on the (disjoint) subdivision W = ⋃n≥−1 Jn ∪ ⋃n≥−1 Jn ∪
{RL∞ }.
Of course, a priori we have to restrict ourselves to W ∖ O. However, one easily verifies
the following lemma.
22 K. SCHEICHER, V.F. SIRVENT, AND P. SURER

Rule ω ∈ W V
1 J−1 W−1 0
2 J−1 W−1 1
3 J0 W0 10P −1
4 Jn (n ≥ 1) Wn 0n−1 1̄10P −2
5 Jn (n ≥ 0) Wn 0n 1̄10P −1
6 RL∞ R 0

Table 3. For ω in the specified set we substitute the initial word W ∈

{L, R}∗ by V ∈ {1̄, 0, 1}∗ . Hence, Φ(ω) can be obtained recursively by con-
catenation, i.e. Φ(ω) = V Φ ○ σ ∣W ∣ (ω).

Lemma 7.4. Let W ∈ {R, L}∗ . Then Φ(W RRL∞ ) = Φ(W LRL∞ ).
Proof. The proof is left to the reader. 
Hence, our definition of Φ coincides with the fact that we identify RRL∞ with LRL∞ .
In other words, it does not make any difference whether we define d as we did in Section 2
or we include α−1 in the right interval and say that S(α−1 ) = R.
Since the order of the rules is not completely arbitrary (for example, Rule 4 can never be
followed by Rule 4) it makes sense to describe Φ by a directed graph. The labels contain
the respecting rules, according to Table 3.

3 4

Jn (n ≥ 1)
1
1 J−1 3
J0 4
1 2
6 3 5
4

{RL∞ } 2
J−1
′ 2
Jn′ (n ≥ 0)
5 5
The graph drafts the possible sequences of rules and with its aid it is easy to verify
that, indeed, Φ(ω) ∈ Ω for all ω ∈ W. However, the graph is far from being a deterministic
automaton. Each of the two vertices on the right represents an infinite number of states
(sets) and edges with the same origin have the same labels.
Theorem 7.5. The function Φ is a measure-theoretically continuous and bijective function
from the binary shift W onto the shift of finite type Ω. For each ω ∈ W there exists positive
integers p = p(ω) and q = q(ω) such that Φ ○ σ p (ω) = σ q ○ Φ(ω). The map Φ cannot be
realised by a finite state transducer.
Proof. By Proposition 2.2, v is continuous on W. The generalised fractional part φ (more
precisely, its restriction to [0, 1]) is continuous on [0, 1] with ξ as only exception, and, by
Proposition 4.6, δ is continuous on I ∖ {x ∈ I ∶ ∃n ≥ 1 ∶ ταn (x) = ξ − 1}. Now observe that
 DYNAMICAL PROPERTIES OF THE TENT MAP 23

φ ○ v maps P ∶= {ω ∈ W ∶ ∃n ≥ 0 ∶ σ n (ω) = R∞ } onto {x ∈ I ∶ ∃n ≥ 1 ∶ ταn (x) = ξ − 1} and v

maps R∞ ∈ P onto ξ. Thus, Φ is continuous on W ∖ P, i.e. almost everywhere.
Actually, it is easy to directly show that Φ is not continuous for x ∈ P by observing that
for each n ≥ 1 we have Jn ∩ Jn′ ∈ P.
Similarly, we can show the measure-theoretical bijectivity for Φ. We know that δ maps
I bijectively onto Ω ∖ U (with U defined as in (4.2)), v maps W ∖ O onto [0, 1] and φ is
bijective if we restrict to [0, 1). Hence, since v(RL∞ ) = 1, we have that Φ maps W ∖ O
bijectively onto Ω ∖ U.
The assertion concerning the integers p and q is clear for W ∖O by Theorem 1.1. However,
one easily verifies that it holds for the elements of O as well.
Finally observe that, for a given ω ∈ W, the identification of the correct J ∈ J such that
ω ∈ J cannot be realised by considering an initial word of finite length, in other words,
we have to “look arbitrary far into the future”. Hence, Φ is not given by a finite state
transducer. 
Finally, we want to note that an “inverse” function Φ′ ∶ Ω Ð→ W can be constructed
quite analogously. The δ-images of the relevant sets In and In′ are, again, not given by
cylinder-sets. More precisely, one has to observe that for x > 0 the first nonzero digit of
δ(x) equals 1 while for x < 0 the first nonzero digit of δ(x) equals 1.

8. Problems and questions

We finally want to state some questions that arise in context with our analysis.
(1) Do the special Pisot numbers α−1 and α−3 satisfy the finiteness property (F)? Com-
puter experiments did not yield any periodic element (except for 0), hence, we
conjecture that (F) holds in these two cases.
(2) For special Pisot numbers the set of purely periodic elements is always finite and
contains 0 (see [6, Theorem 3.2]). Of course, if 0 is the only purely periodic element
then (F) holds. For α = α−2 the set of purely periodic elements consists of 0 and ξ
only (see [6, Corollary 3.2b]). In this context two questions arise. At first, is the set
of purely periodic elements of Tα finite for α = α−6 or α = α6 ? Furthermore, give a
characterisation of the purely periodic points for the cases studied in Proposition 6.2
(different from α−2 ).
(3) In Section 7 we showed that the function Φ is not given by a finite state transducer.
However, theoretical computer science provides a large number of more general
objects. It would be interesting whether Φ fits into any known context.

References
[1] S. Akiyama, T. Borbély, H. Brunotte, A. Pethő, and J. M. Thuswaldner, Generalized
radix representations and dynamical systems. I, Acta Math. Hungar., 108 (2005), pp. 207–238.
[2] B. Derrida, A. Gervois, and Y. Pomeau, Iteration of endomorphisms on the real axis and
representation of numbers, Ann. Inst. H. Poincaré Sect. A (N.S.), 29 (1978), pp. 305–356.
[3] W. M. Y. Goh, Dynamical representation of real numbers and its universality, J. Number Theory,
33 (1989), pp. 334–355.
 24 K. SCHEICHER, V.F. SIRVENT, AND P. SURER

[4] C. Kalle and W. Steiner, Beta-expansions, natural extensions and multiple tilings associated with
Pisot units., Trans. Am. Math. Soc., 364 (2012), pp. 2281–2318.
[5] J. C. Lagarias, H. A. Porta, and K. B. Stolarsky, Asymmetric tent map expansions. I.
Eventually periodic points, J. London Math. Soc. (2), 47 (1993), pp. 542–556.
[6] , Asymmetric tent map expansions. II. Purely periodic points, Illinois J. Math., 38 (1994),
pp. 574–588.
[7] D. Lind and B. Marcus, An introduction to symbolic dynamics and coding, Cambridge University
Press, Cambridge, 1995.
[8] A. Rényi, Representations for real numbers and their ergodic properties, Acta Math. Acad. Sci.
Hungar, 8 (1957), pp. 477–493.
[9] K. Schmidt, On periodic expansions of Pisot numbers and Salem numbers, Bull. London Math. Soc.,
12 (1980), pp. 269–278.
[10] C. J. Smyth, There are only eleven special Pisot numbers, Bull. London Math. Soc., 31 (1999),
pp. 1–5.
[11] P. Surer, ε-shift radix systems and radix representations with shifted digit sets, Publ. Math. (Debre-
cen), 74 (2009), pp. 19–43.
1 Institut für Mathematik, Universität für Bodenkultur, Gregor-Mendel-Straße 33, A-1180 Vienna, Austria
E-mail address: [email protected]
2 Departamento de Matemáticas, Universidad Simón Bolı́var, Apartado 89000, Caracas 1086-A, Venezuela
E-mail address: [email protected]
3 Institut für Mathematik, Universität für Bodenkultur, Gregor-Mendel-Straße 33, A-1180 Vienna, Austria
E-mail address: [email protected]
URL: http://www.palovsky.com

View publication stats