TOKYO J. MATH.

VOL. 12, No. 2, 1989

On the Fractal Curves Induced from the Complex

Radix Expansion

Shunji ITO

Tsuda College

\S 0. Introduction.
Let be a quadratic integer in a complex quadratic field $Z(\sqrt{m}i)$
$\alpha$

and $N(=N(\alpha))$ be the norm of Let be a set of quadratic integer8 $\alpha$

. $\mathcal{D}$

in $Z(\sqrt{m}i)$
whose cardinality is equal to the norm of , and denote it by $\alpha$

$\mathcal{D}=\{\gamma_{0}, \gamma_{1}, \cdots, \gamma_{N-1}\}$

, $r_{i}\in Z(\sqrt{m}i)$
.
A pair $(\alpha, \mathcal{D})$
is called a number system on $Z(\sqrt{m}i)$ if every quadratie
integer $\beta$
in $Z(\sqrt{m}i)$ is uniquely repre8ented in the form
$\beta=r_{0}+r_{1}\alpha+\cdots+\gamma_{j}\alpha^{j}$ , $\gamma_{i}\in \mathcal{D}$ $(0\leqq i\leqq j)$ (0.1)

and we say that is expanded with base and digits

$\beta$
if it $\alpha$ $r_{i}(0\leqq i\leqq j)$

is so represented. Most primitive example of the number system found

in [9] and [10] is as follows: take $\alpha=i-1$ and , then $\mathcal{D}=\{0,1\}$

1) is a number system on Gaussian field $Z(i)$ , and

$(\alpha, \mathcal{D})$

2) the Hausdorff dimension of the boundary of the set

$X_{i-1}=\{\sum_{k=1}^{\infty}a_{k}(i-1)^{-k}|a_{k}e\mathcal{D}\}$

is equal to

$\frac{2\log x}{\log 2}\doteqdot 1.5236$

where is the positive root of

$\lambda$ $\lambda^{3}-\lambda^{2}-2=0$
. This fact is extended as
follows:
THEOREA (Katai-Szabo [8] and Gilbert [7]). Let $\alpha$
be an integer in
$Z(i)$ and take $\mathcal{D}=\{0,1,2, \cdots, N-1\}$ , then

Received April 4, 1988

Revised February 10, 1989
300 SHUNJI ITO

1) $(a, \mathcal{D})$
is a number system if and only if
${\rm Re} a<0$ and ${\rm Im} a=\pm 1$

and
2) the Hausdorf dimension of the boundary of
$X_{-n\pm i}=\{\sum_{k=1}^{\infty}a_{k}(-n\pm i)^{-k}|a_{k}e\mathcal{D}\}$

is equal to
$\frac{2\log_{\lambda_{n}}}{\log(n^{2}+1)}$

where $N_{\mathfrak{n}}$
is the positive root of $x^{\theta}-(2n-1)x^{2}-(n-1)^{2}x-(n^{2}+1)=0$ .
As a generalization of 1) in the theorem, we have
THEOREM 0.1 (Gilbert [6]). Let be a quadrat integer in a complex $a$ $ic$

quadratic field $Z(\sqrt{m}i)$ and be the norm of , and

$N$ $N-1$ }. $a$ $\mathcal{D}=\dagger 0,1,$
$\cdots,$

Then is a number system on $Z(\sqrt mi)$ if and only if

$(a, \mathcal{D})$

$a=-n\pm\sqrt{m}i$ $(n=0,1,2, \cdots)$ if $-m\equiv 2,3$ (mod.4),

$a=-n\pm i$ $(n=1,2, \cdots)$ if $m=1$ ,

and (0.2)

$a=$ $(n=0,1,2, \cdots)$ if $-m\equiv 1$ (mod.4) ,

$a=$ $(n=2,3, \cdots)$ if $m=3$ .

The purpose of this paper is to see that for each base $\alpha$
in Theorem
0.1, the boundary of the set

$X_{\alpha}=\{\sum_{k=1}^{\infty}a_{k}\alpha^{-k}|a_{k}e\{0,1, \cdots, N-1\}\}$

is essentially a fraetal eurve. To state more preeisely, we have the

following result:

RESULT. For each number system in $Z(\sqrt{m}i)$ given by $(a, \mathcal{D})$

Theorem 0.1, a curve satisfying the following property is constructed

$K_{\alpha}$

on a complex plane:
(1) $K_{\alpha}=boundary$
of $X_{\alpha}$
,
 FRACTAL CURVES 301
(2) (space tiling)

$\iota eZ(\overline{n}i)\bigcup_{\sqrt{}}(X_{\alpha}+z)=C$ and int. $(X_{\alpha}+z)\cap int.(X_{\alpha}+z^{\prime})=\emptyset$

(if $z\neq z^{\prime}\in Z(V\overline{m}i)$

),

(3) (self similarity)

$aX_{\alpha}=\bigcup_{j=0}^{N-1}(X_{\alpha}+j)$ ,

(4) int. $X_{\alpha}\ni O$

,
(5) the Hausdorff dimension of the curve $K_{\alpha}$
is equal to
$\frac{2\log_{\lambda_{n}}}{\log N}$

where $N_{n}$
is a positive solution of
$h^{8}-(2n-1)x^{2}-(N-2n)x-N=0$ if $a=-n\pm\sqrt{m}i$

( $-m\equiv 2,3$ (mod.4), $n=1,2,$ $\cdots$ ),

$\lambda^{3}-(2n-2)x^{2}-(N-2n+1)x-N=0$ if $a=\frac{-2n+1\pm\sqrt{m}i}{2}$

( $-m\equiv 1$ (mod.4), $n=1,2,$ $\cdots$ ) ,

$\lambda^{3}-(N-1)\lambda-N=0$
if $a=\frac{1\pm\sqrt{m}i}{2}$

( $-m\equiv 1$ (mod.4)) ,
and
the curve is a rectangle, if $a=\pm\sqrt{m}i$ ( $-m\equiv 2,3$ (mod.4),
$K_{\alpha}$
). $m\geqq 2$

To construct the curves , we consider endomorphisms on the free $K_{\alpha}$ $\theta$

group of rank 2 associated with the number system . This idea $(a, \mathcal{D})$

is essentially that of Dekking [5]. But we cannot apply directly the

Dekking’s method on the endomorphisms , because the endomorphisms $\theta$

have “strong” cancellations. Therefore, we must consider the reduction

to the endomorphisms without caneellation. In fact, in \S 1 we consider
the first reduction of which we call the adjoint $\theta$

with respect to . $\theta_{w}$ $\theta$

As the second step of reduetion, the lifting endomorphism on a free $\Theta$

group of rank 3, whieh has no eancellation, is introdueed in \S 2.

ACKNOWLEDGEMENT. This paper eonsists of the main part of the
results obtained in the seminar theses written by the members of ITO’s
302 SHUNJI ITO

seminars in 1986 and 1987 of Department of Mathematies, Tsuda College.

The members of these seminars are as follows: Y. FUZISAWA,
Y. HANAMURA, K. KASAHARA, M. KISHI, A. KURITA, S. WAKAMEI,
Y. WATANABE, Y. YOSHIYA in 1986 and M. IMAKURA, M. MORISHITA,
M. NAKAYAMA, Y. NAKAYAMA, K. NARA, T. OKAZAKI,
N. TAKEKOSHI in 1987.

\S 1. Endomorphisms associated with number systems.

In this seetion, we introduce a clas8 of endomorphisms on free group
$G\{a, b\}$ assoeiated with number systems $(\alpha, \mathcal{D})$
adopting Dekking’s method
[5].
Let $G\{a, b\}$ be a free group of rank 2, that is, we eonsider $G\{a, b\}$ as
the quotient set of the free semigroup generated by $S:=\{a, b, a^{-1}, b^{-1}\}$ $S^{*}$

where we define the equivalence relation by setting $w\sim v(w, veS^{*})$ iff

$w$ and determine the same words after some eancellations, and we call
$v$

the words in reduced words.

$ S^{*}/\sim$

$\pi:G\{a, b\}\rightarrow Z(\sqrt{m}i)(\subset C)$ be a eanonieal homomorphism, i.e.,

Let $\pi$

is determined by $\pi(a),$ together with the following:

$\pi(b)$

$\pi(w^{-1})=-\pi(w)$ , $\pi(vw)=\pi(v)+\pi(w)$ for $v,$ $weG\{a, b\}$ ,

and the canonical homomorphism associated with base $\pi_{\alpha}$ $a$ is given by
speeifying and as follows: $\pi_{\alpha}(a)$ $\pi_{\alpha}(b)$

DEFINITION 1.1.
$a\rightarrow 1$

$\pi_{\alpha}:b\rightarrow\sqrt{m}i$
if $a=-n+\sqrt{m}i$ $(n=0,1,2, \cdots)$ ,

$\pi_{\alpha}:ab\rightarrow-\sqrt{m}i\rightarrow$
if $a=-n-\sqrt{m}i$ $(n=0,1,2, \cdots)$ ,

$a\rightarrow 1$

if $a=$ $(n=1,2, \cdots)$

$\pi_{\alpha}:b\rightarrow\frac{1+\sqrt{m}i}{2}$

$a\rightarrow 1$

$\pi_{a}:b\rightarrow\frac{1-\sqrt{m}i}{2}$
if $a=$ $(n=1,2, \cdots)$ .

The ease of $\alpha=(1\pm\sqrt{m}i)/2$ will be discussed in Remark 3.6.

Let be an endomorphism of $G\{a, b\}$ , i.e.
$\theta$
is determined by $\theta$
$\theta(a)$ ,
$\theta(b)$
together with the following relations:
 FRACTAL CURVES 303
$\theta(w^{-1})=(\theta(w))^{-1}$ , $\theta(vw)=\theta(v)\theta(w)$ for $v,$ $w\in G$ ,

and the endomorphism $\theta_{\alpha}$

associated with base $\alpha$

is given as follows:
DEFINITION 1.2.
$a\rightarrow a^{N}ba^{-(N+n)}$
$\theta_{\alpha}$
; if $\alpha=-n\pm\sqrt{m}i$ $(n=0,1,2, \cdots)$ ,
$b\rightarrow(a^{n}b^{-1})^{n}a^{-N}$

$\theta_{\alpha}:a\rightarrow a^{N}ba^{-(N+n)}$
if $\alpha=$ $(n=1,2, \cdots)$
$b\rightarrow(a^{n}b^{-1})^{n-1}a^{-N}$

where $a^{n}$
means a string of $n$ conseeutive $a’ s$
.
Let $T_{\alpha}:Z(\sqrt{m}i)\rightarrow Z(\sqrt{m}i)$
be an endomorphism of $Z(\sqrt{m}i)$
defined by
multiplieation by $\alpha$
, that is, is defined by
$T_{\alpha}$

$T_{\alpha}(z)=\alpha z$
.
Then we have the following commutative diagram:
$G\{a, b\}\rightarrow^{\theta_{\alpha}}G\{a, b\}$

$\downarrow\pi_{a}$ $\downarrow\pi_{\alpha}$
(1.1)
$Z(\sqrt{m}i)\rightarrow^{T_{a}}Z(\sqrt{m}i)$

i.e. $T_{\alpha}\circ\pi_{\alpha}=\pi_{\alpha}\circ\theta_{\alpha}$
.
For each base , which assigns polygonal eurves
$\alpha$
, we define a map $K_{\alpha}$

to reduced words as follows: for is the line segment from $(0,0)$ $s\in S,$ $K_{\alpha}[s]$

to $i.e.,$
$\pi_{\alpha}(s),$
, and for $s_{1}\cdots s_{k}eG\{a, b\},$
$K_{\alpha}[s]=\{t\pi_{\alpha}(s);0\leqq t\leqq 1\}$ $K_{\alpha}[s_{1}\cdots s_{k}]$

is the polygon with vertiees $(0,0),$ , i.e., $\pi_{\alpha}(s_{1}),$

$\cdots,$
$\pi_{\alpha}(s_{1}\cdots s_{k})$ $K_{\alpha}[s_{1}\cdots s_{k}]=$

, where $x+A=\{x+y;y\in A\}$ Moreover, if a re-

$U_{j=}^{k}1(\pi_{a}(s_{1}\cdots s_{j-1})+K_{\alpha}[s_{j}])$
.
duced word $w$ satisfies $f(w)=0$ , then the curve is defined by $K_{\alpha}[w]$

$f(v)+K_{\alpha}[w_{1}]$ where is by $w=v\cdot w\cdot v^{-1}$ and is chosen as longest

$w_{1}$
$giv_{1}en-$
$v$

word satisfying $w=v\cdot w_{1}\cdot v$ .

For simplieity we denote sometimes $\pi,$ $\theta,$ $\alpha$
and $K$ instead of $\pi_{\alpha},$
$\theta_{\alpha},$ $T_{\alpha}$

and , respectively.
$K_{a}$

REMARK 1.3. We ean reduee the ease of base $\overline{a}=-n-\sqrt{m}i$

or
$(-2n+1-\sqrt mi)/2$ to the case of base $\alpha=-n+\sqrt{m}i$ or $(-2n+1+\sqrt mi)/2$ .
In fact, from Definitions 1.1 and 1.2, we know
$\theta_{a}^{k}(aba^{-1}b^{-1})=\theta\frac{k}{\alpha}(aba^{-1}b^{-1})$
,
$K[\theta_{\alpha}^{k}(aba^{-1}b^{-1})]=K[\theta\frac{k}{\alpha}(aba^{-1}b^{-1})]$
304 SHUNJI ITO

where 2 is the eonjugate of $zeC$ and $\overline{A}=\{\overline{a}|aeA\}$

. Therefore, we have
$a^{-k}K[\theta_{\alpha}^{\iota}(aba^{-1}b^{-1})]=\overline{a}^{-k}K[\theta\frac{k}{\alpha}(aba^{-1}b^{-1})]$
.
Thus $\overline{a}^{-k}K[\theta\frac{k}{\alpha}(aba^{-1}b^{-1})]$
is obtained by flipping $a^{-k}K[\theta_{\alpha}^{k}(aba^{-1}b^{-1})]$
over the
real axis.
Therefore, we only discuss the case of $a=-n+\sqrt{m}i$ or $a=$
$(-2n+1+\sqrt{m}i)/2$ , from now on.
By the definition of $\theta$

and $K$ , we know that

$K[\theta(aba^{-1}b^{-1})]=\partial\bigcup_{p=0}^{N-1}(F[aba^{-\iota}b^{-\iota}]+p)$ (1.2)

where $F[aba^{-1}b^{-1}]$ is the unit parallelogram whose boundary is $K[aba^{-1}b^{-1}]$ .

The following property is obtained induetively from a geometrie eonsid-
eration.
FUNDAMENTAL PROPERTY 1.4. Let be the domain en- $F[\theta^{k}(aba^{-1}b^{-1})]$

closed by , then the curve and the domain

$K[\theta^{k}(aba^{-1}b^{-\iota})]$ $K[\theta^{k}(aba^{-1}b^{-1})]$

satisfy the following

$F[\theta^{k}(aba^{-1}b^{-1})]$ properties for each : $k$

1) is a simple closed curve.

$K[\theta^{k}(aba^{-\iota}b^{-1})]$

2) (k-step space tiling)

i) $\bigcup_{*z_{(}\Gamma_{n\}}}\{a^{-\iota}F[\theta^{k}(aba^{-1}b^{-1})]+z\}=C$ .
ii) int. $\{a^{-k}F[\theta^{k}(aba^{-1}b^{-1})]+z\}\cap int.\{\alpha^{-k}F[\theta^{k}(aba^{-1}b^{-1})]+z^{\prime}\}=\emptyset$

(if $z\neq z^{\prime}eZ(\sqrt{m}i)$ ).

3) (k-step self similarity)
$a^{-\iota+1}F[\theta^{k}(aba^{-1}b^{-\iota})]=\bigcup_{j=0}^{N-1}(a^{-k+1}F[\theta^{k-1}(aba^{-1}b^{-1})]+j)$

where $aA$ means $\{\alpha z|zeA\}$ .

PROPOSITION 1.5. The following relation holds:
$\alpha^{-k+1}F[\theta^{k}(aba^{-1}b^{-1})]=\bigcup_{*e\Gamma}(a^{-k+1}F[aba^{-1}b^{-1}]+z)$

where $\Gamma=\{\sum_{l=0}^{k-1}a_{l}/a^{l}|a_{l}e\{0,1, \cdots, N-1\}\}$ .

The proof is obtained from 3) of Property 1.4 by induetion.
The purpose of this paper is to show that the limit set of the curve
is the boundary enclosing the set
$a^{-k}K[\theta^{k}(aba^{-1}b^{-1})]$ in Result in \S 0. $X_{\alpha}$

But we eannot apply Dekking’s method directly, beeause the endomorphism

has eancellations. So we try to eonsider the steps of reduction in order
$\theta$

to apply Dekking’s method.

 FRACTAL CURVES 305
First 8tep is to eonstruct an endomorphism related to as follows. $\theta_{w}$ $\theta$

Let be an endomorphism on $G\{a, b\}$ and $w$ be an element of $G\{a, b\}$ ,

$\theta$

then we define an endomorphism by $\theta_{w}$

$\theta_{v}(v)=w^{-1}\theta(v)w$ for $veG\{a, b\}$ .

We call the endomorphism
to a word $w$ .
$\theta$

. an adjoint endomorphism of with respect

$\theta$

DEFINITION 1.6. For each endomorphism $\theta$

in Definition 1.2, we define

the adioint as follows:

$\theta_{w}\ddagger b\rightarrow a^{-(N-n)}b^{-1}(a^{n}b^{-1})^{n-1}a\rightarrow ba^{-n}$

if $\alpha=-n+\sqrt{m}i$ $(n=0,1,2, \cdots)$

where $w$ is chosen as $w=a^{N}$ ,

$\theta_{w}:b\rightarrow a^{-tN-n)}b^{-1}(a^{n}b^{-1})^{n-2}a\rightarrow ba^{-n}$

if $a=$ $(n=1,2, \cdots)$

where $w$ is ehosen as $w=a^{N}$ .

Then we have a proposition:
PROPOSITION 1.7. $\alpha^{-k}K[\theta^{k}(aba^{-1}b^{-1})]$
is congruent to $a^{-k}K[\theta_{w}^{k}(aba^{-1}b^{-1})]$ ,
that is,
$a^{-k}K[\theta^{k}(aba^{-1}b^{-1})]=a^{-k}K[\theta_{w}^{k}(aba^{-1}b^{-1})]+\delta_{k}$ ,
where $\delta_{k}=\sum_{l-1}^{k}\alpha^{-l}\pi(w)$
.
PROOF. From the definition of $\theta_{w}$
, we see the following by induction:
$\theta^{k}(aba^{-1}b^{-1})=w_{k}\theta_{w}^{k}(aba^{-1}b^{-1})w_{k}^{-1}$

where $w_{k}=w\theta_{w}(w)\theta_{w}^{2}(w)\cdots\theta_{w}^{k-1}(w)$ . Therefore,

$K[\theta^{k}(aba^{-1}b^{-1})]=\sum_{l=0}^{k-1}\pi(\theta_{w}{}^{t}(w))+K[\theta_{w}^{k}(aba^{-1}b^{-1})]$
.
Using the relation $\pi\circ\theta_{w}^{k}=a^{k}\circ\pi$
by (1.1), we have
$a^{-k}K[\theta^{k}(aba^{-\iota}b^{-1})]=\sum_{l=1}^{k}a^{-\iota}\pi(w)+a^{-k}K[\theta_{w}^{k}(aba^{-1}b^{-1})]$
.
\S 2. Lifting endomorphism.
In this section, we induce an endomorphism called a lifting of $\theta$

whieh has no caneellation.

306 SHUNJI ITO

Let be a free group on generators $A,$ $B$ and $C$, and define
$G\{A, B, C\}$

a homomorphism $\Phi:G\{A, B, C\}\rightarrow G\{a, b\}$ called a block code map as follows:
DEFINITION 2.1.
$A\rightarrow ba^{-n}$

$\Phi$
: $B\rightarrow a^{-(N-n)}b^{-1}$
if $\alpha=-n+\sqrt{m}i$ $(n=1,2, \ldots)$ ,
$C\rightarrow a^{N}$

$A\rightarrow ba^{-n}$

$\Phi$
: $B\rightarrow a^{-(N-n)}b^{-1}$
if $\alpha=$ $(n=1,2, \cdots)$ .
$C\rightarrow a^{N}$

REMARK 2.2. In the case of $a=\pm\sqrt{m}i$ , the adjoint $\theta_{w}$

in Definition
1.6 has no cancellation, and moreover the relation
$a^{-k}K[\theta_{w}^{k}(aba^{-1}b^{-1})]=K[aba^{-1}b^{-1}]$

holds for all $k$

. Therefore, we need not discuss this ease any further
(see Example 2).

For each we define the endomorphism , which is called alifting-

$\theta$
, $\Theta$

endomorphism of , on the free group $G\{A, B, C\}$ as follows:

$\theta$

DEFINITION 2.3.
$A\rightarrow BA^{-(2n-1)}$

$\Theta$

: $B\rightarrow A^{-(N-2n)}C$
if $\alpha=-n+\sqrt{m}i$ $(n=1,2, \cdots)$ ,
$C\rightarrow A^{N}$

$A\rightarrow BA^{-(2n-2)}$

$\Theta$
: $B\rightarrow A^{-(N-2n+1)}C$ if $a=$ $(n=1,2, \cdots)$ .
$C\rightarrow A^{N}$

This definition is derived from an easy ealculation as follows: in the

case of $\alpha=-n+\sqrt{m}i(n=1,2, \cdots)$ ,

$\Theta(A)=\Phi^{-1}(\theta_{w}(ba^{-n}))$

$=\Phi^{-1}(a^{-(N-n)}b^{-1}(a^{n}b^{-1})^{n-1}(ba^{-n})^{-n})$

$=\Phi^{-1}(a^{-(N-n)}b^{-1}(ba^{-n})^{-(2n-1)})$

$=BA^{-(2n-1)}$

and so on. Therefore we have the following proposition.

 FRACTAL CURVES 307
PROPOSITION 2.4. The endomorphisms $\Theta$

defined above satisfy the

relation:
$\Phi 0\Theta(S)=\theta_{w}\circ\Phi(S)$
for $Se\{A^{\pm 1}, B^{f1}, C^{\pm 1}\}$ .
LEMMA 2.5. $\Theta^{k}(ABC)$ has no canceuation for any $k$
.
PROOF. Let us denote
$\Theta^{k}(ABC)=A_{1}^{(k)}A_{2}^{(k)}\cdots A_{*(k)}^{(k)}$ $(k=1,2, \cdots)$ ,

and we call the pairs of alphabets $A_{j}^{(k)}A_{\dot{g}+1}^{(k)}(j=1,2, \cdots, \epsilon(k)-1)$ and

admi ssible pairs in $\Theta^{k}(ABC)$ . We consider the set
$A_{(k)}^{(k)}A_{1}^{(k)}$ $\ovalbox{\tt\small REJECT}_{k}(k=$

$0,1,2,$ ) of all admissible pairs in $\Theta^{k}(ABC)$ , and put

$\cdots$ $\ovalbox{\tt\small REJECT}_{\Theta}=\cup\ovalbox{\tt\small REJECT}_{k}$
. Then
we obtain
$\ovalbox{\tt\small REJECT}_{\Theta}=\{AB,$
$BC,$ $CA,$ $BA^{-1},$ $A^{-1}A^{-1},$ $A^{-1}C,$ $AA,$ $AB^{-1},$ $B^{-1}A$ ,
$A^{-1}B,$ $A^{-1}C^{-1},$ $C^{-\iota}A,$ $B^{-1}A^{-\iota}$
}. (2.1)

In fact, in the ease of $\alpha=-n+\sqrt{m}i(n=1,2, \cdots)$ , we know

$\mathscr{A}_{0}=\{AB, BC, CA\}$ .
From the relation that $\Theta(AB)=BA^{-(2n-1)}A^{-(N-2n)}C$ , we see that the ad-
missible pairs appearing in $\Theta(AB)$ are $BA^{-1},$ and $A^{-I}C$. After $A^{-I}A^{-I}$

similar considerations for the pairs $BC$ and $CA$ , we obtain

$\mathscr{A}_{1}=$
{ $BA^{-1},$ $A^{-1}A^{-1},$ $A^{-1}C,$ $CA,$ $AA,$ AB}.
Continue this procedure till $k=3$ , then we see . In the $\ovalbox{\tt\small REJECT}_{s}=\ovalbox{\tt\small REJECT}_{k}(k>3)$

case of $\alpha=(-2n+1+\sqrt{m}i)/2(n=1,2, \cdots)$ , we obtain (2.1) in the same

manner. For each pair , it is easy to see that the cancellation $S\cdot Te\ovalbox{\tt\small REJECT}_{\Theta}$

does not oecur in , that is, the endomorphism has no can-

$\Theta(S)\cdot\Theta(T)$ $\theta$

eellation on for any . $\Theta^{k}(ABC)$ $k$

Therefore, we have the following proposition:

PROPOSITION 2.6. Let be the lifting endomorphism of given by $\Theta$ $\theta$

Definition 2.3, then the endomorphism has no cancellation and has $\Theta$ $\Phi$

no cancellation on $\Theta^{k}(ABC)$ and moreover the following relation holds:

$\theta_{w}^{k+1}(aba^{-1}b^{-1})=\Phi 0\Theta^{k}(ABC)$ .
PROOF. The fact that has no cancellation on $\Theta^{k}(ABC)$ is discussed $\Theta$

in Lemma 2.5. On the other hand, from Definition 2.1, it is easy to

see the cancellation does not occur in for any admissible pair $\Phi(S)\cdot\Phi(T)$
308 SHUNJI ITO

Therefore
$ S\cdot Tes\swarrow\Theta$
. $\Phi$
has no cancellation on $\Theta^{k}(ABC)$ . From Definition
2.1, we know
$\theta_{w}(aba^{\leftarrow 1}b^{-\iota})=\Phi(ABC)$

and from Proposition 2.4, we obtain inductively

$\theta_{w}^{k+1}(aba^{-1}b^{-1})=\Phi\cdot\Theta^{k}(ABC)$ .
\S 3. Fractal curves induced from endomorphisms.
In this seetion, we see the existence of a limit set of $\alpha^{-k}K[\theta^{k}(aba^{-1}b^{-1})]$

which is in faet a ”fractal curve” except when $\alpha=\pm\sqrt{m}i$ .

Let us define a map as follows: $\hat{\pi}:G\{ABC\}\rightarrow C$

$\hat{\pi}:=\pi\circ\Phi$
. (3.1)

Then, we have the following proposition.

PROPOSITION 3.1. For each $\Theta$

defined in Definition 2.3, the following

diagram $ commute\epsilon$ :
$\Theta$

$G\{A, B, C\}\leftrightarrow G(A,$ $B,$ $C$ }

$\downarrow\hat{\pi}$ $\downarrow\hat{\pi}$

$\rightarrow^{T_{\alpha}}$

$C$ $C$

i.e. $\tau_{\alpha^{\circ\hat{\pi}=\hat{\pi}0}}e$
.
PROOF. This follows from
$\Phi\circ\Theta=\theta_{w}\circ\Phi$
and $\pi\circ\theta_{v}=\pi\circ\theta$
.
In fact,
$\hat{\pi}0\Theta=\pi\circ\Phi\circ\Theta=\pi\circ\theta_{\tau v}\circ\Phi=\pi\circ\theta\circ\Phi=T_{\alpha}\circ\pi 0\Phi=T_{\alpha}\circ\hat{\pi}$
.
We define a map , which assigns polygonal curves on $\hat{K}$
$C$
to reduced
words of free group $G\{A, B, C\}$ of rank 3, as follows:
$\hat{K}[A_{1}A_{2}\cdots A.]:=K[\Phi(A_{1}A_{2}\cdots A.)]$ for $ W=A_{1}A_{2}\cdots$
A. $eG\{A, B, C\}$ . (3.2)

Then, from Proposition 2.6 and (3.2), we see

PROPOSITION 3.2. For each $\Theta$

in Definition 2.3, we have

$\hat{K}[\Theta^{k}(ABC)]=K[\theta_{w}^{k+1}(aba^{-1}b^{-1})]$
.
 FRACTAL CURVES 309
we show there exists a limit set of
Now, $\alpha^{-k}K[\theta_{w}^{k}(aba^{-1}b^{-1})]$
as a fractal
curve. Let us define a non-negative matrix $N_{e}=(a_{ST})(S, Te\{A, B, C\})$ ;

$a_{sr}=the$ number of $S$

or $S^{-1}$
appearing in $\Theta(T)$ .
From Definition 2.3, the matrix $N_{e}$
is given explicitly by

$N_{e}=\left(\begin{array}{lll}2n-1 & N-2n & N\\1 & 0 & 0\\0 & 1 & 0\end{array}\right)$
if $\alpha=-n+\sqrt{m}i$ $(n=1,2, \cdots)$

(3.3)

$N_{\Theta}=\left(\begin{array}{lll}2n-2 & N-2n+1 & N\\1 & 0 & 0\\0 & 1 & 0\end{array}\right)$
if $a=$ $(n=1,2, \cdots)$ .
We know the matrix $N_{\theta}$
is aperiodie, that is, there exists $n$ such that
$N_{e^{n}}>0$ .
Therefore, by Perron-Frobeniuv’ theorem, there exists an eigen row vector
$x=(x_{A}, x_{B}, x_{\sigma})$
with respect to the maximum eigenvalue $\lambda_{\Theta}(>1)$
of $N_{\Theta}$

satisfying the condition:

$x_{A}+x_{B}+x_{C}=1$ , $x_{A},$ $x_{B},$ $x_{G}>0$ .
Let us define the partition $\xi_{k}=\{I(A^{(k)});1\leqq i\leqq j(k)\}$ of interval $I=[0,1]$
associated with inductively as follows:
$\Theta^{k}(ABC)=A_{1}^{(k)}A_{2}^{(k)}\cdots A_{jtk)}^{(k)}$

(i) The partition is given by $\xi_{0}$

$\xi_{0}=\{I(A_{1}^{(0)}), I(A_{2}^{(0)}), I(A_{\theta}^{(0)})\}$

where $(A_{1}^{(0)}, A_{a}^{(0)}, A_{\epsilon}^{(0)})=(A, B, C)$

and eaeh $I(A_{i}^{(0)})(i=1,2,3)$ is an interval
whose length is equal to $x_{A_{i}^{(0)}}$ , namely, $I(A_{1}^{(0)})=[0, X_{A_{1}^{(0)}}],$ $I(A_{2}^{(0)})=[X_{A_{1}^{(0)}}$ ,
$X_{A_{1}^{(0)}}+X_{A_{2}^{(0)}}],$ $I(A_{3}^{(0)})=[X_{A_{1}^{(0)}}+X_{A_{2}^{(0)}}, X_{A_{1}^{(0)}}+X_{A_{2}^{(0)}}+X_{4_{S}^{(0)}}]$ .
(ii) If the partition is given, the partition is construeted $\xi_{k-1}$ $\xi_{k}$

by partitioning the interval $I(A^{(k-1)})$ of according to $\Theta(A^{(k-1)})=$ $\xi_{k-1}$

$B_{1}B_{2}\cdots B_{(k-1,)}$ in the ratio : : : . $x_{B_{1}}$ $x_{B_{2}}$

$\cdots$
$x_{B_{(k-1,)}}$

In view of the relation

$x_{B_{1}}+x_{B_{2}}+\cdots+x_{B_{(k-1,i)}}.=x_{\Theta}x_{A_{i}^{(k-1)}}$ ,
we obtain the following proposition:
PROPOSITION 3.3. For the partition $\xi_{k}=\{I(A_{\dot{f}}^{(k)});\Theta^{k}(ABC)=A_{1}^{(k)}\cdots A_{\dot{g}(k)}^{\{k)}\}$

of , the length of intervals are estimated uniformly as

$I$
310 SHUNJI ITO

$|IA_{j}^{(k\}})|\sim\frac{1}{|x_{\theta}|^{k}}$ .
Let us define a polygonal map $\psi_{k}:I\rightarrow C$ mapping for each $k$
sub-
interval as follows:
$I(A_{j}^{(k)})$

$\psi_{k}(I(A_{j}^{(k)}))=a^{-(k+1)\{\sum_{h=1}^{\dot{g}-1}(\hat{\pi}(A_{h}^{(k\}}))+\hat{K}[A_{j}^{(k)}]\}}$ for each $I(A_{j}^{(k)})e\xi_{k}$ ,

then we see
$\psi_{k}(I)=a^{-(k+1)}\hat{K}[\Theta^{k}(ABC)]$ . (3.4)

From the definition of and , we know the end points of , whieh

$\hat{K}$ $\Theta$
$\hat{K}[S]$

are given by and $\pi(S)$ , coincide with the end points of

$0$ for $a^{-1}\hat{K}[\Theta(S)]$

.
all $Se\{A^{f1}, B^{f1}, C^{f1}\}$ From Proposition 2.6, the endomorphism and $\Theta$ $\Phi$

have no cancellation, and from Proposition 3.3, we have

$d(\psi_{k}(I), \psi_{k+\iota}(I))\leqq d_{0}\cdot x_{\Theta}^{-k}$ (3.5)

where $d(\cdot, )$ is the Hausdorff metric on a family of compact subset of

$C$ and $d_{0}$
is given by
$d_{0}=\max_{Se|4,B,CI}d(\hat{K}(S), \alpha^{-1}\hat{K}[\Theta(S)])$ .
Therefore, by (3.4) and (3.5) we have
PROPOSITION 3.4. Let be the limit $\psi$
of the curves $\psi_{k}$
. Then $\psi$
is a
continuous closed curve and satisfies
$\psi(I)=\lim_{k\rightarrow\infty}a^{-k}K[\theta_{w}^{k}(aba^{-1}b^{-1})]$ .
PROOF. From (3.5), is well defined as a eontinuous closed curve. $\psi$

From (3.4) and Proposition 3.2, the set is char- $\lim_{k\rightarrow\infty}a^{-k}K[\theta_{w}^{k}(aba^{-1}b^{-1})]$

acterized as the image of by $I$ $\psi$

Now we state our theorem:

THEOREM 3.5. Let be a base of number system on $Z(\sqrt{m}i)$ and $a$
$\theta$

be the endomorphism associated with the base Then there exists a $a$ .
curve as the limit of
$\psi_{\alpha}:I\rightarrow C$ Put be a closed set $a^{-k}K[\theta^{k}(aba^{-1}b^{-1})]$ . $F_{\alpha}$

enclosed by , then satisfies the following

$\psi_{\alpha}$
condition: $F_{\alpha}$

1) (space tiling)

$\bigcup_{**z_{(\sqrt{}}\overline{n}i)}(F_{\alpha}+z)=C$ and int. $(F_{\alpha}+z)\cap int.(F_{\alpha}+z^{\prime})=\emptyset$

(if $z\neq z’\in Z(\sqrt{m}i)$ ),

 FRACTAL CURVES 311
2) (self similarity)

$\alpha F_{\alpha}=\bigcap_{j=0}(F_{\alpha}+j)N1$ ,

3) int. $F_{\alpha}\ni O$
.
PROOF. By Proposition 1.7, we know that the set $\alpha^{-k}K[\theta^{k}(aba^{-1}b^{-1})]$

is congruent to and converges. Therefore,

$\alpha^{-k}K[\theta_{w}^{k}(aba^{-1}b^{-1})]$ $\sum_{l=1}^{\infty}\alpha^{-l}\pi(w)$

by Proposition 3.4, there exists a eurve as the limit of $\psi_{\alpha}\ddagger I\rightarrow C$

. Now, since for each , Fundamental Property 1.4 is

$\alpha^{-k}K[\theta^{k}(aba^{-1}b^{-1})]$ $k$

satisfied, we see by taking the limit as that the eonclusions 1) $ k\rightarrow\infty$

and 2) of the theorem are valid. (In the case of $\alpha=\pm\sqrt{m}i$ , we have
the conelusions 1) and 2) from Proposition 1.7 and Remark 2.2 directly.)
For the statement 3), we know by Fundamental Property 1.4, 3)
that
$F[\theta^{k}(aba^{-1}b^{-1})]=\bigcup_{je\Gamma_{k}}\{F[aba^{-1}b^{-1}]+j\}$

where $\Gamma_{k}=\{r_{0}+\gamma_{1}\alpha+\cdots+r_{k-1}\alpha^{k-1};0\leqq r_{i}\leqq N-1\}$ . We observe that if we

put , then
$\alpha^{*}=\sum_{k=1}^{\infty}(N-1)|\alpha|^{-k}$
is greater than a diameter of the set $2\alpha^{*}$

. From Theorem 0.1, there exists $k_{0}\in N$ and

$F_{\alpha}$ $\alpha_{j(k_{0})}eZ(\sqrt{m}i)$ $\alpha_{0},$ $a_{1},$ $\cdots,$

such that
1) $F[\theta^{k_{0}}(aba^{-1}b^{-1})]\supset\bigcup_{1\leqq i\leq j(k_{0})}(F[aba^{-1}b^{-1}]+a_{i})\supset\partial F[\theta^{k_{0}}(aba^{-1}b^{-1})]$

and
2) $\min_{1\leqq i\leq j(k_{0})}|\alpha_{i}|\geqq 2\alpha^{*}$ .
Relation 1) can be extended inductively as follows:
1)’ $ F[\theta^{k_{0}+n}(aba^{-1}b^{-1})]\supset$ $\cup$ $(F[\theta^{n}(aba^{-1}b^{-1})]+\alpha^{n}\alpha_{l})$

$1\leq i\leqq j(k_{0})$

$\supset\partial F[\theta^{k_{0}+n}(aba^{-1}b^{-1})]$
.
Divide the relation 1)’ by $\alpha^{n}$
and let $n$ tend to infinity, then we have
$\alpha^{k_{0}}F_{\alpha}\supset\bigcup_{1\leqq i\leq j(k_{0})}(F_{\alpha}+\alpha_{i})\supset\alpha^{k_{0}}K_{a}$ .
Therefore, the distance of $\alpha^{k_{0}}K_{\alpha}$
from the origin is estimated by
$d(a^{k_{0}}K_{\alpha}, 0)\geqq\min_{1\leq i\leq\dot{g}(k_{0})}d(F_{\alpha}+\alpha_{i}, 0)\geqq\min_{1\leq i\lessgtr j(k_{0})}|\alpha_{i}|-a^{*}\geqq\alpha^{*}$ .
This is equivalent to saying that int. $(F_{\alpha})\ni O$
.
REMARK 3.6. In the case of $n=0$ in (0.2) we eonsider the boundary
of the set $X_{\alpha}=\{\sum_{k=1}^{\infty}a_{k}\alpha^{-k}|a_{k}\in\{0,1, \cdots, N-1\}\}$ for $\alpha=(1\pm\sqrt{m}i)/2$ . It
3I2 SHUNJI ITO

i8 easy to see that the sets and $X_{(1+\Gamma_{n)/2}}$ $X_{t1-\Gamma_{ni)/g}}$

are eongruent to
and , which
$X_{(-1-,*)/2}\sqrt{}^{-}$
is the case of
$X_{(-1-\Gamma_{l1i)/2}}$ $n=1$ in $(0,2)$ , respeetively.
More precisely, we know

$X_{\alpha}=X_{-\alpha}-\sum_{k=1}^{\infty}(N-1)(-a)^{-(2k-1)}$ .
Therefore, the shape of the boundary of $X_{\alpha}$
is reduced to that of $X_{-\alpha}$
.
If we want to find out how we ean construct the boundary of $X_{\alpha}$

$(a=(1\pm\sqrt{m}i)/2)$ directly, we need a somewhat more complicated proeedure

as follows: let us define the canonical homomorphism associated with $\pi_{\alpha}$

base by $a$

$a\rightarrow 1$

$\pi_{\alpha}$
: if $a=\frac{1\pm\sqrt{m}i}{2}$

$b\rightarrow\frac{1\mp\sqrt{m}i}{2}$

and the endomorphism $\theta_{\alpha}$

assoeiated with base $\alpha$

by
$a\rightarrow ab^{-\iota}$

:
.
$\theta_{\alpha}$

$b\rightarrow ba^{N}b^{-1}$

For $\theta_{a}$
, consider the adjoint $\theta_{w}$
sueh that
$a\rightarrow b^{-1}a$

$\theta_{w}$
: $b\rightarrow a^{N}$

and define the bloek code map $\Phi:G\{a, \beta, \gamma\}\rightarrow G\{a, b\}$ such that
$a\rightarrow b^{-1}$

$\Phi$
: $\beta\rightarrow a$

$\gamma\rightarrow a^{N-1}$

and define the blocks $A,$ $B$ and $C$

such that
$ A:=a\beta$ ,
$B:=\gamma a^{-1}$ ,
$C:=\beta^{-1}\gamma^{-1}$ ,
and define lifting endomorphisms $\Theta$

and $\hat{\Theta}$

such that
$a\rightarrow\beta^{-1}\gamma^{-1}$

$\Theta$
: $\beta\rightarrow a\beta$

$\gamma-(a\beta)^{N-1}$ ,
 FRACTAL CURVES 313
$\alpha\rightarrow\gamma^{-1}$

$\hat{\Theta}$

: $\beta\rightarrow\alpha$

$\gamma\rightarrow\beta(\alpha\beta)^{N-1}$
,
then we see that
(1) $\Phi(ABC)=\theta_{w}(aba^{-J}b^{-1})$ ,
(2) $\Phi\Theta^{k}(ABC)=\theta_{w}^{k}(aba^{-1}b^{-1})$ ,
(3) $\beta^{-1}\hat{\Theta}(S)\beta=\Theta(S)$
for $se\{A^{\pm 1}, B^{\pm 1}, C^{\pm 1}\}$ ,
(4) has no cancellation,
$\hat{\Theta}(ABC)$

has no eancellation on
(5) $\Phi$ $\hat{\Theta}^{k}(ABC)$
.
Using this fact, we have
$\alpha^{-(k+1)}K[\theta_{w}^{k+1}(aba^{-1}b^{-1})]=\alpha^{-(k+1)}f(\Phi(v_{n}))+\alpha^{-(k+1)}K[\Phi(\hat{\Theta}(ABC))]$

where . Therefore the limit set of

$v_{n}=\beta\Theta(\beta)\cdots\Theta^{k-1}(\beta)$ $a^{-(k+1)}K[\theta^{k+1}(aba^{-1}b^{-1})]$

is characterized by the limit set of $\alpha^{-(k+1)}K[\Phi(\hat{\Theta}(ABC))]$

. (For details, see
Ito-Ohtsuki [11].)

\S 4. Hausdorff dimension.
In this section, the Hausdorff dimension of the curve $K_{\alpha}$
is caleulated
by using Frostman’s lemma.

LEMMA (Frostman (cf. [1], [3])). If there exists a measure $\mu$

on a
set $X$ satisfying
$\mu(B)\leqq c\cdot|B|$
for any ball $B$ , (4.1)

where $|B|$ is the radius of a ball B. Then the Hausdorf dimension of

$X$ is estimated as

$\dim_{H}(X)\geqq s$ .
For eaeh base $a$ except $\alpha=\pm\sqrt{m}i$
, we induce the measure $\mu_{\alpha}$
on
$K_{\alpha}$
by
$\mu_{\alpha}=(\psi_{\alpha})_{*}\circ x$
(4.2)

where $\lambda$
is the Lebesgue measure on , then we see $I$

Put $s=2\log x_{\Theta}/\log N$, then the measure

PROPOSITION 4.1. $\mu_{\alpha}$

satisfies
the assumption (4.1) in Frostman’s lemma.
PROOF. Let $B_{r}$
be a ball with radius $r$ , then we have from (4.2)
314 SHUNJI ITO

$\mu(B_{r})=\sum_{)j:^{\psi(I(1_{j}^{(k)}}4)\cap B_{r}\neq\emptyset}\mu(\psi_{\alpha}(I(A_{j}^{(k)}))\cap B_{r})$

for all where is as defined in Proposition 3.3. From Proposition

$k$ $I(A_{\dot{f}}^{(k)})$

3.3 there exists a constant , independent of , sueh that $c$ $k$

$\mu(B_{r})\leqq c\cdot x_{\theta}^{-k}Card\{I(A_{\dot{f}}^{(k)});\psi_{\alpha}(I(A_{j}^{(k)}))\cap B_{r}\neq\emptyset\}$ (4.3)

for all $k$

. Now, we choose $k$
so as to satisfy
$|a|^{-k}\leqq r<|\alpha|^{-k+1}$ ,

then we can rewrite (4.3) to read:

$\mu(B_{r})\leqq c\cdot r^{2\log\lambda_{\Theta}/\log N}\cdot Card\{I(A_{j}^{\langle k)});\psi_{\alpha}(I(A_{\dot{f}}^{(k)}))\cap B_{r}\neq\emptyset\}$
.
We note that the cardinality of is smaller than $\{I(A_{\dot{f}}^{(k)});\psi_{\alpha}(I(A_{j}^{(k)}))\cap B_{r}\neq\emptyset\}$

that of . On the other hand, the curves

$\{(I(A_{\dot{f}}^{(k)});a^{k-1}\cdot\psi_{\alpha}(I(A_{\dot{f}}^{(k)}))\cap B_{1}\neq\emptyset\}$

joint two points which are neighbouring points of the

$a^{k-1}\cdot\psi_{\alpha}(I(A_{j}^{(k)}))$

lattiee generated by { on . Therefore, the cardinality of $\hat{\pi}(A),\hat{\pi}\{B)\}$ $R^{2}$

is uniformly bounded, and so the inequality

$\{I(A_{\dot{f}}^{(k)});a^{k-1}\cdot\psi_{\alpha}(I(A_{j}^{(k)}))\cap B_{1}\neq\emptyset\}$

holds.
THEOREM 4.2. For each base $a$ , the Hausdorf dimension of $K_{a}$
is
given by

$\dim_{H}K_{\alpha}=\frac{2\log_{\lambda_{\Theta}}}{\log N}$

where $\lambda_{\theta}$
is a positive solution of
$x^{3}-(2n-1)x^{2}-(N-2n)x-N=0$ if $\alpha=-n\pm\sqrt{m}i$

( $-m\equiv 2,3$ (mod.4), $n=1,2,$ $\cdots$ ) ,

$\lambda^{3}-(2n-2)x^{2}-(N-2n+1)\lambda-N=0$ $’\dot{b}f$ $a=\frac{-2n+1\pm\sqrt{m}i}{2}$

( $-m\equiv 1$ (mod.4), $n=1,2,$ $\cdots$ ) ,

and

$x^{3}-(N-1)x-N=0$ if $a=\frac{1\pm\sqrt{m}i}{2}$

( $-m\equiv 1$ (mod.4)).

PROOF. By Proposition 4.1, we have the estimation from below:

 FRACTAL CURVES 315

$\dim_{H}K_{\alpha}\geqq 2\cdot\frac{\log_{\lambda_{\Theta}}}{\log N}$

.
The opposite inequality is obtained by the growth rate $\lambda_{\theta}^{n}$
of the
length and contracting constant
$\theta^{n}(aba^{-1}b^{-1})$
of the map $|\alpha||$
$T_{\alpha}$
(see [2],
[3]).

\S 5. Examples.
EXAMPLE 1. The simplest base in $Z(i)$ is knowp to be $a=\dashv 1\pm i$ ,
which has the smallest norm among $\alpha=-n\pm i$ . On the base $\alpha=-1+i$ ,
the canonical homomorphism and the endomorphism is given by $\pi$ $\theta$

$a\rightarrow 1$ $a\rightarrow a^{2}ba^{-3}$

$\pi$
: :
.
$\theta$

$b\rightarrow i$ $b\rightarrow ab^{-1}a^{-2}$

Take the adjoint $\theta_{w}$

with $w=a^{2}$ , and eonsider the block code map $\Phi$
:
$A\rightarrow ba^{-1}$
$a\rightarrow ba^{-1}$

$\theta_{w}$
: $b\rightarrow a^{-1}b^{-1}$
$\Phi$
: $B\rightarrow a^{-1}b^{-1}$

$C\rightarrow a^{2}$
,
then the lifting endomorphism $\Theta$

is given by
$A\rightarrow BA^{-1}$

$\Theta$

: $B\rightarrow C$

$C\rightarrow A^{2}$
.
The curves $a^{-k}K[\theta^{k}(aba^{-1}b^{-1})](k=1,2, \cdots)$ are obtained as in Figure 1.
In Figure 2, we see the manner of
$a^{-k}K[\theta_{w}^{k}(aba^{-1}b^{-1})]$

and the block code This set is known as the skin of twindragon
$\Phi$
.
(cf. Dekking [4] and Sandelbrot [10]).
The shapes in the case of $\alpha=-n+i(n=2,3)$ are seen in Figure 3
(cf. Gilbert [7]).

EXAMPLE 2. In $Z(\sqrt{2}i)$ , the base is given by $a=-n\pm\sqrt{2}i$ . $\alpha$

In
the case of $a=\sqrt{2}i$ , the endomorphism and are given by $\theta$
$\theta_{w}$

$a\rightarrow a^{2}ba^{-2}$ $a\rightarrow b$

$\theta$

: $b\rightarrow a^{-2}$
and $\theta_{w}$
: $b\rightarrow a^{-2}$
316 SHUNJI ITO

FIGURE (1) $\alpha=-1+i,$ $a^{-k}K[\theta^{k}(aba^{-1}b^{-1})]$ $(k=1,2, \cdots)$

 FRACTAL CURVES 317

FIGURE (2) $\alpha=-1+i,$ $\alpha^{-k}K[\theta_{w^{k}}(aba^{-1}b^{-1})]$ $(k=1,2, \cdots)$

318 SHUNJI ITO

$a=-3+i,$ $\dim_{H}K=$ 1.5496

FIGURE (3)

$a=\frac{-3+\sqrt{}\overline{3}i}{2},$
$\dim_{H}K=$ 1.6575 $a=\frac{-5+\sqrt{}\overline{3}i}{2}$
, dim $K_{H}=$ 1.5981
FIGURE (5)
 FRACTAL CURVES 319

FIGURE (6) $\alpha=\frac{-1+\sqrt{7}i}{2},$

$\dim_{H}K=$ 1.2107
320 SHUNJI ITO

where $w=a^{2}$ . The eurves are eongruent for all

$a^{-k}K[\theta^{k}(aba^{-1}b^{-1})]$ $k$
and
the limit set is a reetangle. In the ease of , the eurves $n\neq 0$ $K_{-nf^{\sqrt{}}\overline{2}i}$
are
fraetal (see Figure 4).
EXAMPLE 3. In $Z(\sqrt{3}i)$ , the base is given by $a=(-2n+1\pm\sqrt{}\overline{3}i)/2$
$a$

$(n=2,3, \cdots)$ . The shapes of are given in Figure 5. $K_{\alpha}$

EXAMPLE 4. In $Z(\sqrt{7}i),$ $a=(\pm 1\pm\sqrt{7}i)/2$ are basis with digits $\{0,1\}$ .

In the case of $a=(-1+V\overline{7}i)/2$ , the eanonieal homomorphism and the $\pi$

endomorphi8m is given by $\theta$

$\pi$
: $a\rightarrow 1b\rightarrow(1+\sqrt{7}i)/2$ $\theta$

: $b\rightarrow a^{-2}a\rightarrow a^{2}ba^{-\epsilon}$

The curves $a^{-k}K[\theta^{k}(aba^{-\iota}b^{-1})](k=1,2, \cdots)$ are obtained as in Figure

6. By Remark 1.3, we know that the curve is obtained by $K_{(-1-\Gamma_{7})/2}$

flipping the curve over the real axis. And by Remark 3.6, the
$K_{(-1+\Gamma_{7)/2}}$

curves and are obtained by the parallel displacement

$K_{(1+\$\overline{7}i)/2}$ $K_{(1-\Gamma\tau)/2}$

of and
$K_{(-1-\Gamma_{7})/2}$ respectively. $K_{(-1+\Gamma_{7i)/2}}$

References
[1] T. J. BEDFORD, Ph. D. Thesis, Warwick Univ. (1984).
[2] T. J. BEDFORD, Dimension and dynamics for fractal recurrent sets, preprint, King’s
College Research Centre, Cambridge (1985).
[3] F. M. DEKKING, Recurrent Sets: Fractal Formalism, Reports Dept. Math. Inf., no. 82-32
(1982), Delft.
[4] F. M. DEKKING, Recurrent sets, Adv. Math., 44 (1982), 78-104.
[5] F. M. DEKKING, Replicating superfigures and endomorphisms of free groups, J. Combin.
Theory Ser. A, 32 (1982), 315-320.
[6] W. J. GILBERT, Radix representations of quadratic fields, J. Math. Anal. Appl., 83 (1981),
264-274.
[7] W. J. GILBERT, The fractal dimension of sets derived from complex basis, to appear in
Canad. Math. Bull.
[8] I. KATAI and J. SZABO, Canonical number sy8tems for complex integers, Acta Sci. Math.
(Szeged), 37 (1974), 225-260.
[9] D. E. KNUTH, The Art of Computer Programming vol. 2, Seminumerical Algorithms,
Addison-Wesley (1981).
[10] B. B. MANDELBROT, The Fractal Geometry of Nature, Freeman (1983).
[11] Sh. ITO and M. OHTSUKI, On the fractal curves induced from endomorphisms on 2 states
free group, preprint.

Present Address:
DEPARTMENT OF MATHEMATICS, TSUDA COLLEGE
TSUDA-MACHI, KODAIRA, TOKYO 187, JAPAN