Topological properties of Zp and Qp and

Euclidean models
Samuel Trautwein, Esther Röder, Giorgio Barozzi
November 13, 2011

1 Topology of Qp vs Topology of R
Both R and Qp are normed fields and complete metric spaces, both are
completions of Q. Since Q is dense in both of them, they are seperable.
An open ball in Qp with center a and radius r is denoted by

B(a,r) := {x ∈ Qp | ka − xkp < r}

since
{kx − ykp | x, y ∈ Qp } = {pn | n ∈ Z} ∪ {0}
we only need to consider the balls of radi r = pn , where n ∈ Z.
The sphere with center a and radius r is denoted by

S(a,r) = {x ∈ Qp | kx − akp = r}

Proposition 1.1. The Sphere S(a,r) is an open set in Qp .

Proof. Let x ∈ S(a,r) , choose ε < r. We now show that B(x,ε) ⊂ S(a,r) .
Let y ∈ B(x,ε) ⇒ kx − ykp < kx − akp = r and by Proposition 1.15 (Katok)
follows ky − akp = kx − akp = r which means that y ∈ S(a,r) .
Since x and y were arbitrary it follows that
[
S(a,r) = B(x,ε)
x∈S(a,r)

Therefore the sphere S(a,r) is a union of open sets, so it is open itself.

1
Proposition 1.2. Open balls in Qp are open and closed.
Proof. B(a,r) is closed ⇔ B(a,r) c = {x ∈ Qp | kx − akp ≥ r} is open. We know

B(a,r) c = S(a,r) ∪ D

where D := {x ∈ Qp | kx − akp > r}

Because of the last Proposition it sufficies to show that D is open.
So, let y ∈ D, ky − akp =: r1 > r We claim: B(y,r1 −r) ⊂ D.
Otherwise there exists an x ∈ B(y,r1 −r) such that kx − akp ≤ r but using the
triangle inequality, leads to

r1 = ky − akp = ky − x + x − akp ≤ ky − xkp + kx − akp < r + (r1 − r) = r

which is a contradiction to our assumption

S r1 < r.
Therefore, B(y,r1 −r) ⊂ D, so D = y∈D B(y,r1 −r) which means that D is
open.
Hence, since B(a,r) is closed, the open balls in Qp have no boundary and
in particular S(a,r) is not the boundary of B(a,r) . In addition we get that

B (a,pn ) 6= B(a,pn ) = B(a,pn )

in fact we have

B (a,pn ) = {x ∈ Qp | kx − akp ≤ pn } = {x ∈ Qp | kx − akp < pn+1 } = B(a,pn+1 )

Proposition 1.3. Every point of a ball is its center, i.e.

∀b ∈ B(a,r) we have B(a,r) = B(b,r)

Proposition 1.4. Two balls in Qp have a non empty intersection if and only
if one is contained in the other, i.e.

B(a,r) ∩ B(b,s) 6= ∅ ⇔ B(a,r) ⊂ B(b,s) or B(b,s) ⊂ B(a,r)

Proof. ⇐ is clear
⇒ let y ∈ B(a,r) ∩ B(b,s) w.l.o.g assume r ≤ s, by the Proposition before we
have
B(a,r) = B(y,r) ⊂ B(y,s) = B(b,s)

2
 Z_7

B(1,1)

B(6,1) B(2,1)
B(0,1)

B(5,1) B(3,1)
B(4,1)

Zoom into B(1,1)

B
B(8,1/7)
B

B(43,1/7) B(15,1/7)
B(1,1/7)

B(36,1/7) B(22,1/7)

B(29,1/7)

Zoom into B(1,1/7)

B(50,1/49)

B(99,1/49)
B(295,1/49)

B(1,1/49)

B(246,1/49) B(148,1/49)

B(197,1/49)

Figure 1: Z7

3
 S
Taking Z7 as example we see that S(0,1) = x∈{1,...,6} B(x,1)
in addition it holds that

∀x, y ∈ Z7 s.t. x ∈ y + 7k Z7 we have B(x, 7−(k−1) ) = B(y, 7−(k−1) )

let x ∈ Z7 be arbitrary, k ∈ N then we can observe that

[
B(x,7−k ) = B(x + j ∗ 7k+1 , 7−(k+1) )
j∈{0,...6}

We can generalize this : S

Let p be a prime number, then we have that S(0,1) = x∈{1,...,p−1} B(x,1) in
addition we have that

∀x, y ∈ Zp s.t. x ∈ y + 7k Zp we have B(x, p−(k−1) ) = B(y, p−(k−1) )

and for an arbitrary element x ∈ Zp , for all k ∈ N it holds that

[
B(x,p−k ) = B(x + j ∗ pk+1 , p−(k+1) )
j∈{0,...p−1}

Proposition 1.5. The sphere S(a,r) is open and closed

Proof. We have already shown that every sphere is open. We observe that
S(a,r) = B (a,r) ∩ B(a,r) c which is closed because it is a finite intersection of
closed subsets.

Proposition 1.6. The set of all balls in Qp is countable.

Proof. Let B(a,r) be an arbitrary ball in Qp . We know that r =Pp−s for some s ∈
∞ n
Psa ∈ Qp nthere exist m ∈ (Z) s.t. am 6= 0 and a = n=m a−s
Z. Since n p . Let
a0 := n=m an p , obviously a0 ∈ Q and we have ka − a0 kp < p which
means that a0 ∈ B(a,p−s ) . As before, we have B(a,r) = B(a0 ,p−s ) .
Therefore both, the set of radii and the set of centers of balls in Qp , are
countable which leads to the fact that the set of balls in Qp is countable.

Theorem 1.7. The set Zp is compact and the space Qp is locally compact.

Proof. We know that Zp is sequentially compact, since it is a metric space it

is therefore compact. Because Zp = B (0,1) = B(0,p) it follows that every ball
in Qp is compact. So Qp is a locally compact space.

4
Theorem 1.8. N is dense in Zp .
Proof. let x = . . . a2 a1 a0 ∈ Zp for all n ∈ N define
n
X
xn := . . . 00an an−1 . . . a0 = ai p i ∈ N
i=0

we obtain kx − xn kp < p−n

Theorem 1.9. The space Qp is totally disconnected.
Proof. We show that for an arbitrary element a ∈ Qp the connected com-
ponent Ca of a is equal to {a}. Let a be arbitrary and suppose Ca ) {a}
therefore there exist n ∈ N such that B(a,p−n ) ∩ Ca 6= Ca . But then we have

Ca = (B(a,p−n ) ∩ Ca ) ∪ ((Qp \ B(a,p−n ) ) ∩ Ca )

which is the disjoint union of two open subsets. Therefore Ca is not con-
nected, which is a contradiction.

2 Cantor Set Models of Zp

We start with a repetition of the classical Cantor set C ⊂ [0, 1] and explain
its basic properties. We will observe that Zp is homeomorphic to C for
every prime number p. In the case p = 2 there is a natural homeomorphism
Z2 ∼= C using the triadic expansion of real numbers. In the case p > 2 there
is a natural homeomorphism C (p) ∼ = Zp , where C (p) ⊂ [0, 1] is obtained by a
similar construction as the one of the classical Cantor set C.
This allows us to reduce the initial claim that every Zp is homeomorphic
to C to the statement that C (p) and C are homeomorphic.
S
Definition 2.1 (Cantor Set). Let A = k∈Z [2k, 2k + 1] and C0 := [0, 1]. We
define by induction
Cn := Cn−1 ∩ 3−n A .

Then the Cantor set C is given by

\
C := Cn
n≥0

5
 Each set Cn consists of 2n closed intervals of length 3−n and Cn+1 is
obtained from Cn by removing the middle third in of each of these intervals.
Lemma 2.1 (Properties of the Cantor Set C). The Cantor set C ⊂ [0, 1]
satisfies the following:

1. C is compact.
2. The Cantor set has vanishing Lebesgue measure, i.e. |C| = 0 where | · |
denotes the Lebesgue measure.
3. The Cantor set is perfect (see definition below).
4. The Cantor set is uncountable.

Definition 2.2 (Perfect Set). A closed set E is called perfect, if for every
x ∈ E there exists a sequence (xn ) ⊂ E − {x} converging to x.
Proof. C is clearly bounded and closed, since each Cn is closed, and therefore
C is compact.
Since Cn is the union of 2n intervals of length 3−n we have |Cn | = (2/3)n
and since Cn ⊃ Cn+1 for all n we conclude
 n
2
|C| = lim |Cn | = lim =0
n→∞ n→∞ 3
For the third statement note that ∂Cn ⊂ C. Now pick x ∈ C arbitrary.
We can choose xn ∈ ∂Cn −{x} such that |xn −x| ≤ 3−n and therefore xn → x.
We show that the third statement implies the last one. So let E ⊂ R be
any nonempty perfect subset and assume E = {ei }∞ i=1 is countable. Define
En := E − {en }. Choose x1 ∈ E1 and let I1 be a finite open interval such
that x1 ∈ I1 and e1 ∈/ I 1 . Since E is perfect we have I1 ∩ E2 6= ∅. Then pick
x2 ∈ I2 ∩E2 and let I2 be an open interval such that x2 ∈ I2 ⊂ I1 and e2 ∈/ I 2.
Continuing in this way we obtain a decreasing sequence (In ) of intervals such
that en ∈
/ I n . On the other hand we observe
\
E ∩ I n 6= ∅
n≥1

since all sets E ∩ I n are compact and nonempty. This contradicts our as-
sumption that E is countable.

6
Definition 2.3 (d-Adic Expansion). Let d ∈ Z+ and x ∈ [0, 1]. A d-adic
expansion of x is given by
∞
X
x= xn d−n
n=1

where xn ∈ {0, 1, . . . , d − 1}. It is easy to check that every x ∈ [0, 1] admits

a unique d-adic expansion, when we enforce the digit sequence {xk } to have
not finite support. We denote the d-adic expansion by [x]d = 0, x1 x2 . . ..
For d = 10 this leads to the usual decimal expansion of real numbers and
we have for example 0.1 = 0.0999999 . . .. By definition [1/10]10 is the second
variant.
Lemma 2.2 (Characterization of C via triadic expansion). Let x ∈ [0, 1]
and [x]3 = 0, x1 x2 . . . be the triadic expansion of x. Then it holds x ∈ C if
and only if xn ∈ {0, 2} for every n.
Proof. By induction on n, it follows immediately from the definitions that
x ∈ Cn if and only if x1 , . . . , xn ∈ {0, 2}.
Theorem 2.3 (Homeomorphism of Z2 and C). Z2 equipped with the 2-adic
norm | · |2 is homeomorphic to C equipped with the absolute value as norm.
An explicit homeomorphism is given by
∞
X ∞
X
Φ : Z2 −→ C, n
an 2 7→ (2an )3−(n+1) .
n=0 n=0

xn 2n , y = yn 2n ∈ Z2 . Then it
P P
Proof. Clearly, Φ is bijective. Let x =
holds

|x − y|2 ≤ 2−k ⇔ xn = yn for n ≤ k

⇔ The first k digits in [Φ(x)]3 and [Φ(y)]3 are equal
⇔ |Φ(x) − Φ(y)| ≤ 3−k

This shows that Φ is a homeomorphism.

Now we will extend our discussion to the case of Zp for a prime number
p > 2. The results will follow completely analog to the case p = 2 and
therefore our presentation will be briefer.

7
Definition 2.4 (Cantor Set C (p) ). We define a variant of the classical Cantor
S (p)
set. Let p be a prime number, A = k∈Z [2k, 2k + 1] and C0 := [0, 1]. We
define inductively
(p)
Cn(p) := Cn−1 ∩ (2p − 1)−n A

and \
C (p) := Cn(p) .
n≥0

(p)
The set Cn consists of [(2p − 1)/2]n disjoint open intervals of length
(p)
(2p − 1)−n and Cn+1 is obtained by subdividing each of these intervals into
2p − 1 equal subintervals and then deleting every second open interval.

Lemma 2.4 (Properties of C (p) ). The Cantor set C (p) satisfies the following
properties:

1. C (p) is a compact and perfect subset of the real line with vanishing
Lebesgue measure. In particular C (p) is uncountable.

2. C (p) contains all x ∈ [0, 1] whose (2p − 1)-adic expansion [x]2p−1 =

0, x1 x2 x3 . . . contains only even digits {xn }.

Proof. The proof is completely analogue to the case p = 2.

Theorem 2.5 (Homeomorphism of Zp and C (p) ). Zp equipped with the p-adic

norm | · |p is homeomorphic to C (p) equipped with the absolute value as norm.
An explicit homeomorphism is given by
∞
X ∞
X
Φp : Zp −→ C, an pn 7→ (2an )(2p − 1)−(n+1) .
n=0 n=0

Proof.
P Thenproof Pis the same as for p = 2. It is clear that Φp is bijective. Let
x= xn p , y = yn pn ∈ Zp . Then it holds

|x − y|2 ≤ p−k ⇔ xn = yn for n ≤ k

⇔ The first k digits in [Φ(x)]2p−1 and [Φ(y)]2p−1 are equal
⇔ |Φ(x) − Φ(y)| ≤ (2p − 1)−k

This shows that Φp is a homeomorphism.

8
Theorem 2.6 (Topological equivalence of the spaces Zp ). The spaces Z2 and
Zp are homeomorphic
Recall that all spaces Zp are totally disconnected. Consequently, the
classical Cantor set C and its variants C (p) are totally disconnected and
therefore the theorem follows from the following Lemma.
Lemma 2.7. Any compact perfect totally disconnected subset E of the real
line is homeomorphic to the Cantor set C
Proof. Let m := inf E and M := sup E. We are going to define a function
F : [m, M ] → [0, 1] such that F maps E homeomorphic on C. We will con-
struct this function on the complements [m, M ]\E → [0, 1]\C first, since the
complements are both countable unions of open intervals and therefore much
less complicated then the sets E and C themselves. Since both complements
are dense we can extend this map by continuity to a map F : [m, M ] → [0, 1].
The connected subsets of the real line are precisely the intervals. There-
fore [m, M ]\E is the disjoint union of infinite but countably many open
intervals (they are the connected components of [m, M ]\E) and the same
is true for [0, 1]\C. Let I be the collection of the intervals whose union is
[m, M ]\E and J be the collection whose union is [0, 1]\C. We start with the
construction of an appropriate bijection

Θ : I → J.

Let I1 ∈ I be an interval of maximal length and define Θ(I1 ) = (1/3, 2/3).

Next choose intervals I21 and I22 to the left and right of I1 such that they
have maximal length and define Θ(I21 ) = (1/9, 2/9) and Θ(I22 ) = (7/9, 8/9).
Continuing this process defines Θ on the whole set I, since I contains only
finitely many sets of length greater then some fixed  > 0 and since any two
intervals in I or in J have different endpoints (as E and C are perfect). It
is clear from the construction that Θ is bijective and order preserving in the
sense that if I is to the left of I 0 , then Θ(I) is to the left of Θ(I 0 ).
Now define F as follows: For I ∈ I let F |I : I → Θ(I) be the unique
linear increasing map, which maps I bijectively onto Θ(I). Since E and C are
totally disconnected, they are nowhere dense and thus there exists at most
one continuation F : [m, M ] → [0, 1]. Now it follows from our construction
of Θ that a well-defined continuation F : [m, M ] → [0, 1] is given by

F (x) = sup{ F (y) : y ∈

/ E, y ≤ x }

9
 Let f := F |E . Then f : E → C is a monotone increasing, continuous
bijection and we need to show that g := f −1 is continuous. Note that g is
again monotone increasing. Let x ∈ C and xn → x converge to x. This
sequence contains a monotone subsequence and thus we may assume wlog
that the sequence xn itself is monotone increasing. Clearly,

y := lim g(xn ) = sup g(xn ) ≤ g(x)

n→∞ n≥1

Now assume y < g(x). Since E is closed, we have y ∈ E and g −1 (y) < x.
This implies that y < xn for large n and by monotonicity y < g(xn ). This
contradicts the definition of y and shows continuity of g.

3 Euclidean models of Zp
3.1 Linear models of Zp
Let us choose b ∈ (1, ∞) and and use it as numeration base in [0, 1], i.e. we
write ∞
a0 a1 X ai
[0, 1] 3 a = + 2 + ... =
b b i=0
bi+1
with 0 ≤ ai < b, ai ∈ N.
We consider
∞ ∞
X
i
X ai
ψ := ψb,p : Zp → [0, 1] ai p →
7 α (1)
i=0 i=0
bi+1

with α ∈ R+ chosen such that maxζ∈Zp ψ(ζ) = 1.

Proposition 3.1. Let ψ : Zp → [0, 1] be defined as in (1) with b > 1 and let
p be a prime number. Then

1.
b−1
α=
p−1
2. ψ and its inverse are always continuous;

10
 3. ψ is injective if b > p (and in this case it is a homeomorphism onto its
image ψ(Zp )).
Proof. 1. Since the ai ’s are all positive numbers, maxζ∈Zp ψ(ζ) is attained
whenPthese are maximal (thus ai = p − 1 for all i ∈ N). So take
ζ= ∞ i
i=0 (p − 1)p = −1. Then
∞  
X 1 1 p−1
1 = α(p − 1) i+1
= α(p − 1) 1 −1 =α
i=0
b 1− b b−1

and thus the claim follows.

P∞
2. ThisP
is analogous to the proof of Theorem 2.6. Let x = n=0 xn pi and
y= ∞ i
n=0 yn p ∈ Zp . Then

|x − y|p ≤ p−k ⇔ xn = yn for n ≤ k

⇔ The first k digits in [ψ(x)]b and [ψ(y)]b are equal
⇔ |ψ(x) − ψ(y)| ≤ b−k
and since both b, p ≥ 1 we have always continuity for both maps.
if i>k (p − 1)pi , pk ∈ Zp have distinct images
P
3. We will have injectivity
in [0, 1] with ψ( i>k (p − 1)pi ) < ψ(pk ).
P

∞ ∞ k+2
X b−1 X p−1 X 1 X1
ψ( (p − 1)pi ) = i+1
= (b − 1)( i
− i
)=
i>k
p − 1 i=k+1
b i=0
b i=0
b
1 
1 − bk+2

1
= (b − 1) 1 − =
1− b 1 − 1b
b bk+2 − 1 b
= (b − 1) (1 − k+2
) = k+2 = b−k−1
b−1 b b
b − 1 1 b − 1
ψ(pk ) = = b−k−1
p − 1 bk+1 p−1
b−1
Thus we must have p−1
> 1, i.e. b > p.

Remark. As seen in the previous section, the Cantor set C p corresponds to

ψ2p−1,p .
When b > p, ψ is thus an homeomorphism and we get a subset of [0, 1]
homeomorphic to Zp .

11
Definition 3.1. A subset A of Rn homeomorphic to Zp is called Euclidean
model of Zp . In the special case n = 1, we call such subsets linear models
of Zp .
Definition 3.2. A fractal is a self-similar geometric object. In other words,
it can be split into parts, each of which is a reduced-size copy of the whole.
Definition 3.3. Informally speaking, the self-similarity dimension of a
fractal is the statistical quantity that indicates how it appears to fill the
space. More explicitely, if we take an object with euclidean dimension d and
reduce its linear size by 1l in each spacial dimension, then it takes N = ld
self-similar objects to cover the original one. Thus
ln(N )
d = logl (N ) = .
ln(l)
Remark. Let us obtain this definition a little bit more formally: Let A be a
d-dimensional fractal and let E(A) denote its extent in the space (which is
intuitively clear). Then E(A) is a homogeneous function of degree d and A
is the union of N translates of 1l A. Thus we get
1 N
E(A) = N · E( A) = d E(A).
l l
Remark. 1. In the case of a linear model of Zp (i.e. when b > p), we
have de facto a fractal with l = b and N = p. As one can imagine,
0 < d = ln(p)
ln(b)
< 1.
2. Note that
b&p ⇒ d % 1.
Example 3.1. 1. The self-similarity dimension is a generalization of the
usual dimension! Let A for example be a cube. If we reduce its linear
size by 1l in each spacial dimension then we obviously need l3 little
cubes to cover the original one. Thus
ln(l3 )
d= = 3.
ln(l)

2. Let us take the Cantor set C (p) as in the previous section. Then
ln(p)
d= .
ln(2p − 1)

12
 Note in particular that the usual Cantor set C = C (2) has dimension

ln(2)
d= = 0.63.
ln(3)

3.2 Higher-dimensional Euclidean models of Zp

We want to generalize the map ψ as follows: in order to obtain an image
in a general space V , we first need to map the digits of a p-adic number to
some points in the space. Now the generalization becomes quite clear, since
we will map this points in V under the map ψ.
In this subsection we will see how α and b has to be chosen to get a good
generalization Ψ of the map ψ. But first let us define rigorously the map Ψ:
Definition 3.4. 1. A finitely dimensional inner product space V over R is
called Euclidean space.

2. Let V be an Euclidean space and let ν : S := {0, 1, . . . , p − 1} ,→ V be

an injective map. Then define V ⊃ Σ := ν(S) and
∞ ∞
X
i
X ν(ai )
Ψ := Ψν,b,p : Zp → V ai p 7→ α (2)
i=0 i=0
bi+1

We already know we can express Zp in a disjoint union as following:

]
Zp = a + pZp .
a∈S

Let us see what is the image of Zp under Ψ:

] [ [ αv 1
Ψ(Zp ) = Ψ( a + pZp ) = Ψ(a) + Ψ(pZp ) = + Ψ(Zp ) (3)
a∈S a∈S v∈Σ
b b

since multiply by p corresponds (under Ψ) to move the dot on the right in the
numeration base b. Note the union is no more necessarily disjoint, but for b
enough large it will be. In this case Ψ(Zp ) is injective and hence a homeo-
morphism (since continuity of Ψ and its inverse are obvious). Furthermore it
is the union of disjoint self-similar images (i.e. a fractal) and hence we have
an iterative construction of the spatial model.

13
Remark. As in the case for linear models, we need a good choice for α, in the
sense that we want the smallest α such that all elements in Σ are reached by
Ψ(Zp ).
Lemma 3.2. Choosing α = b − 1, Ψ(Zp ) remains in the convex hull Σ̄ of Σ
and hence it is the desired α.
Proof. Let λ : V → R be an affine linear functional such that

λ ≤ 1 on Σ and λ(v) = 1 for some v ∈ Σ.

Then ∞ ∞
X ν(ai ) X 1
λ(α )≤α =1
i=0
bi+1 i=0
b i+1

and thus F := Ψ(Zp ) ⊂ Σ̄.

To sum up, we obtain a good generalization Ψ of the homeomorphic map
ψ if we choose α = b − 1 and b large enough in order to get a disjoint union
in (3).
Before going to some clarifying examples let us see how we can obtain an
iterative construction of a fractal through removing pieces:
Remark. The fractal F := Ψ(Zp ) is the intersection of a decreasing sequence
of compact subsets Kn :
Let K0 := Σ̄. Then
[ v F [ v K0
F = α + ⊂ α + =: K1
v∈Σ
b b v∈Σ
b b

and inductively we have

[ v Kn−1
F ⊂ Kn := α + for all n
v∈Σ
b b

This leads to the representation

∞
\
F = Kn
n=1

which is a lot useful for constructing inductively a fractal such as the Sier-
pinski gasket or for us an Euclidean model:

14
 Figure 2: Model of Z3 and Sierpinski gasket

Example√3.2 (Sierpinski gasket). Let p = 3, E = R2 with basis e1 = (1, 0),

e2 = ( 21 , 23 ), b > 1, α = b − 1 (by the previous remark) and define

 0 if k = 0
ν(k) = e1 if k = 1
e2 if k = 2


Then ∞ ∞
X
i
X ν(ai )
Ψ = Ψ3,b,ν : Z3 → V ai 3 7→ (b − 1)
i=0 i=0
bi+1
We observe that the self-similar images are disjoint when b > 2. Thus in that
case we get homeomorphic models of Z3 inside the triangle.
If b = 2 the image is the well-known Sierpinski gasket which is a connected
figure (synonym of not injectivity of Ψ). For example we see that − 23 and 1
∞
have the same image under Ψ: first note − 32 = 1−31
− 1 = i=1 3i . Hence
P

∞  
3 X e1 1 1 e1
Ψ(− ) = i
= e1 1 −1− = = Ψ(1)
2 i=2
2 1− 2
2 2

In general, taking the enumeration base at the ’limit’ of injectivity gives

connected fractals which are the well-know ones!

15
 Figure 3: Model of Z5 and limiting fractal

Example 3.3. Let p = 5, E = R2 and define



 (0, 0) if k =0
 (1, 0) if k =1


ν(k) = (0, 1) if k =2
(−1, 0) if k =3




(0, −1) if k =4


We want to discuss injectivity: because of the symmetry it is enough to check

it for the points lying between (0, 0) P
and (1, 0). We want to avoid that the
two p-adic integers i=1 5 and 1 + ∞
P∞ i i
i=1 3 · 5 have the same image. It is
easy to see this happens when b = 3 since the two vertices of the squares
meet if Ψ(1) = b−1 b
= 23 . As above the limit case b = 3 gives a connected
fractal but not a homeomorphic model of Z5 .
Example 3.4 (Generalization of Example 3.3). Let p > 3, E = R2 and α =
b − 1. We can generate a regular (p − 1)-gon centered in 0 with vertices
e1 , . . . , ep−1 . Let us choose them with norm equal 1 and define ν(k) = ek
for k 6= 0 and ν(0) = 0. Then the injectivity of the map Ψ is obtained by b
large enough: more precisely, for p ≥ 7 the minimal distance between points
in (Ψ(k))k∈S is between Ψ(k) and Ψ(k + 1), k 6∈ 0, p − 1, and it is given by

16
2(b−1) π
b
sin( p−1 ). Then b has to satisfy
∞
X 1 1 b−1 π
(b − 1) i
= < sin( )
i=2
b b b p − 1

and this gives the criterium b > sin( 1 π ) + 1 for injectivity. Note that p = 5
p−1
has the same b of p = 7, since the minimal distance between points in Σ is
between the center 0 and a vertice ek and for p = 7 we get an exagon where
the minimal distance between points in Σ is both the distance between the
center 0 and a vertice ek and the one between two vertices next to each other.
Remark. Let us now calculate the dimension of the fractals obtained by let-
ting b to the limit of injectivity of the map Ψ we have seen in this subsection.
The dimension of the Sierpinski gasket is
ln(3)
d= = 1.58,
ln(2)
the one of the limiting quadratic fractal in Example 3.3 is
ln(5)
d= = 1.46
ln(3)

and the dimension of the fractal in the (p − 1)-gon is

 ln(7)
 ln(3) = 1.77
 if p = 5
ln(p) ! if p ≥ 7
d=
 ln 1
+1

sin( p−1
π
)
e.g. for p = 83 d = 1.339.

3.3 Euclidean models of Qp

We know that we can express Qp as
[
Qp = p−m Zp . (4)
m≥0

But in the previous section we saw that a model of pZp is a contraction of

ratio 1b of the model Zp . So conversely a dilatation of ratio b of the model Zp

17
gives us a model of p1 Zp . Thus inductively by (4) an euclidean model of Qp is
nothing else but an extension of the model of Zp to the whole Euclidean space
V . As one can imagine, the homeomorphism giving the Euclidean model is
∞ ∞
X
i
X ν(ai )
Ψ := Ψν,b,p : Qp → V ai p →
7 α (5)
i=−∞ i=−∞
bi+1

with the same assumptions as for equation (2).

Let us explain this more carefully for the model constructed in Example 3.4:
take the (p − 1)-gon which contains the fractal of Zp and copy it to its (p − 1)
vertices such that the two opposite vertices touch each other; we thus obtain
a new much bigger (p−1)-gon containing p fractals and this will be the model
of p1 Zp = {ζ ∈ Qp | a−k = 0 for k ≥ 2}. Inductively, take the (p − 1)-gon
1
containing the model of pn−1 Zp , copy it to its (p − 1) vertices and obtain a
1
model of pn Zp .

18
Figure 4: Iterative construction of the model of Q7 given the one of Z7

19