This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Hausdorff Dimension of closure of cycles in dd-maps on the circle

Nicholas Payne Northeastern University, Boston, MA 02115, USA [email protected]  and  Mrudul Thatte Columbia University, New York, NY 10027, USA [email protected]
Abstract.

We study the dynamics of the map xdx(mod 1)x\mapsto dx(\text{mod }1) on the unit circle. We characterize the invariant finite subsets of this map which are called cycles and are graded by their degrees. By looking at the combinatorial properties of the base-d expansion of the elements in the cycles, we prove a conjecture of Curt McMullen that the Hausdorff dimension of the closure of degree-m cycles is equal to logm/logd\log m/\log d.

1. Introduction

Degree-dd maps on the unit disk have many interesting geometric, topological and analytic properties, which are closely related to hyperbolic geometry. The dynamics of these maps has an important application in classifying dynamical systems generated by polynomials in single complex variable  [goldberg1992fixed] and it provides useful information about Julia sets and Mandelbrot sets [keller2000invariant, douady1984etude]. To get an understanding of this family of maps, people studied the behaviour of these maps restricted to the boundary unit circle [goldberg1992fixed, goldberg1993fixed]. In the paper [mcmullen2010dynamics], McMullen gave a complete description of simple (i.e. degree 1) cycles of these boundary maps, showing that those simple cycles are an analogue of simple closed geodesics on hyperbolic surfaces in the sense that the closure of union of simple cycles has Hausdorff Dimension 0 and the closure of union of simple closed geodesics has Hausdorff Dimension 1.

Here we consider degree-dd holomorphic maps from the unit disk onto itself. We focus on one such map zzdz\mapsto z^{d}. The restriction of this map to the boundary circle is equivalent to the map xdx(mod 1)x\mapsto dx\ \text{(mod 1)} on /.\mathbb{R}/\mathbb{Z}. Combinatorial features of this map has been studied recently in [PZ, zakeri2021cyclic, tan2022counting]. In this paper, we study cycles of higher degrees in this map, and in particular, calculate the Hausdorff dimension of their closure, which confirms a conjecture of McMullen.

Definition 1 (dd-map).

Let dd\in\mathbb{N}. We define a dd-map to be the map on the unit circle S1=/S^{1}=\mathbb{R}/\mathbb{Z} given by

(1) xdx(mod 1),xS1.x\mapsto dx\ \text{(mod 1)},\forall x\in S^{1}.

In other words, if the base-dd expansion of a point is (0.b1b2b3.)d(0.b_{1}b_{2}b_{3}....)_{d}, then dd-map takes it to the point whose base-dd expansion is (0.b2b3b4.)d(0.b_{2}b_{3}b_{4}....)_{d}.

We say that a dd-map has degree dd.

Definition 2 (Cycle).

Let dd be a positive integer greater than 1. A finite set CS1C\subset S^{1} is called a cycle for the dd-map if and only if the dd-map restricted to CC is a transitive permutation. In terms of the base-dd expansion, CC is given by

(2) C={(0.bibi+1.bnb1b2.bi1¯)d|1in},C=\{(0.\overline{b_{i}b_{i+1}....b_{n}b_{1}b_{2}....b_{i-1}})_{d}|1\leq i\leq n\},

where nn is the size of the cycle and b1,b2,,bnb_{1},b_{2},...,b_{n} are fixed digits in {0,1,,d1}\{0,1,...,d-1\}.

Here are some examples of cycles. Consider the case d=2d=2. Here, {0}={(0.0¯)2}\{0\}=\{(0.\overline{0})_{2}\} is the only 1-element cycle and {13,23}={(0.01¯)2,(0.10¯)2}\{\frac{1}{3},\frac{2}{3}\}=\{(0.\overline{01})_{2},(0.\overline{10})_{2}\} is the only 2-element cycle. For an integer n>2n>2, there are multiple nn-element cycles. For example, there are two 3-element cycles: {17,27,47}\{\frac{1}{7},\frac{2}{7},\frac{4}{7}\} and {37,57,67}\{\frac{3}{7},\frac{5}{7},\frac{6}{7}\}.

Now we define an important invariant of a cycle, called degree.

Definition 3 (Degree of a cycle).

Let dd be a positive integer greater than 1. Let CC be a cycle for dd-map. The degree of CC is the smallest non-negative integer mm for which there exists a degree-mm map f:S1S1f:S^{1}\rightarrow S^{1} such that ff and the dd-map agree on CC. It is denoted by deg(C)deg(C).

Remark 1.

Note that 0deg(C)d0\leq deg(C)\leq d by definition.

Remark 2.

The cycles which contain only one element are fixed points of dd-map and have degree 0. All other cycles have positive degree.

Now, we give two examples.

Example 1.

Consider d=2d=2 and C={37,57,67}C=\{\frac{3}{7},\frac{5}{7},\frac{6}{7}\}. Under the 2-map,

3767,5737,6757.\frac{3}{7}\mapsto\frac{6}{7},\ \frac{5}{7}\mapsto\frac{3}{7},\ \frac{6}{7}\mapsto\frac{5}{7}.

Here the cyclic order of the points in CC is preserved. As we jump on the points in the cycle from 375767\frac{3}{7}\rightarrow\frac{5}{7}\rightarrow\frac{6}{7} (jump over 0) and back to 37\frac{3}{7}, we complete one full circle. As we jump on the images of these points, 67\frac{6}{7}\rightarrow (jump over 0 to) 3757\frac{3}{7}\rightarrow\frac{5}{7} and back to 67\frac{6}{7}, again we complete one circle. This means that the degree of CC is 1. We can show this by explicitly constructing a suitable map f:S1S1f:S^{1}\to S^{1} as follows:

f(x)={x+37if x[37,47]3(x47)if x[47,57]2x1if x[57,67]57+14(x67)if x[67,107]f(x)=\left\{\begin{array}[]{l l}x+\frac{3}{7}&\quad\text{if $x\in\left[\frac{3}{7},\frac{4}{7}\right]$}\\ &\\ 3\left(x-\frac{4}{7}\right)&\quad\text{if $x\in\left[\frac{4}{7},\frac{5}{7}\right]$}\\ &\\ 2x-1&\quad\text{if $x\in\left[\frac{5}{7},\frac{6}{7}\right]$}\\ &\\ \frac{5}{7}+\frac{1}{4}\left(x-\frac{6}{7}\right)&\quad\text{if $x\in\left[\frac{6}{7},\frac{10}{7}\right]$}\\ \end{array}\right.

Observe that ff agrees with the 2-map, and it is of degree 1. As there are no maps with smaller degree, deg(C)=1deg(C)=1.

Example 2.

Consider d=3d=3 and C={15,25,35,45}C=\{\frac{1}{5},\frac{2}{5},\frac{3}{5},\frac{4}{5}\}. Under the 3-map,

1535,2515,3545,4525.\frac{1}{5}\mapsto\frac{3}{5},\ \frac{2}{5}\mapsto\frac{1}{5},\ \frac{3}{5}\mapsto\frac{4}{5},\ \frac{4}{5}\mapsto\frac{2}{5}.

So, as we jump on the points in the cycle from 15253545\frac{1}{5}\rightarrow\frac{2}{5}\rightarrow\frac{3}{5}\rightarrow\frac{4}{5} (jump over 0) and back to 15\frac{1}{5}, we complete one full circle. As we jump on the images of these points, 35\frac{3}{5}\rightarrow (jump over 0 to) 1545\frac{1}{5}\rightarrow\frac{4}{5}\rightarrow (jump over 0 to) 25\frac{2}{5} and back to 35\frac{3}{5}, we complete two circles. So, the degree of CC is 2. We can show this by explicitly constructing a suitable map f:S1S1f:S^{1}\to S^{1} as follows:

f(x)={35+4(x15)if x[15,310]2(x310)if x[310,25]3x1if x[25,35]45+2(x35)if x[35,710]4(x710)if x[710,45]25+12(x45)if x[45,65]f(x)=\left\{\begin{array}[]{l l}\frac{3}{5}+4(x-\frac{1}{5})&\quad\text{if $x\in\left[\frac{1}{5},\frac{3}{10}\right]$}\\ &\\ 2(x-\frac{3}{10})&\quad\text{if $x\in\left[\frac{3}{10},\frac{2}{5}\right]$}\\ &\\ 3x-1&\quad\text{if $x\in\left[\frac{2}{5},\frac{3}{5}\right]$}\\ &\\ \frac{4}{5}+2(x-\frac{3}{5})&\quad\text{if $x\in\left[\frac{3}{5},\frac{7}{10}\right]$}\\ &\\ 4(x-\frac{7}{10})&\quad\text{if $x\in\left[\frac{7}{10},\frac{4}{5}\right]$}\\ &\\ \frac{2}{5}+\frac{1}{2}\left(x-\frac{4}{5}\right)&\quad\text{if $x\in\left[\frac{4}{5},\frac{6}{5}\right]$}\\ \end{array}\right.

Observe that ff agrees with the 3-map, and it is of degree 2. As there are no maps with smaller degree, deg(C)=2deg(C)=2.

Definition 4 (Closure of cycles).

Let m,dm,d\in\mathbb{N} with 1<d1<d and 1md1\leq m\leq d. We define Em,dE_{m,d} as the closure of the union of degree-mm cycles for dd-map.

(3) Em,d={cS1|c is in a cycle of degree-m for the d-map}¯E_{m,d}=\overline{\{c\in S^{1}|c\text{ is in a cycle of degree-$m$ for the $d$-map}\}}

Curtis McMullen discussed in  [mcmullen2010dynamics] the simple (i.e. degree 1) cycles for the dd-map and and computed the Hausdorff Dimension of their closure E1,dE_{1,d}. Recall that Hausdorff Dimension is defined as follows.

Definition 5 (Hausdorff Dimension).

The Hausdorff Dimension of a set EE is defined as

(4) dimH(E)=inf{δ0|inf{riδ|EB(xi,ri)}=0}.dim_{H}(E)=\inf\ \{\delta\geq 0|\ \inf\ \{\sum{r_{i}^{\delta}|E\subset\bigcup{B(x_{i},r_{i})}}\}=0\}.
Theorem 1 (McMullen).

Let dd be a positive integer greater than 1. Then dimH(E1,d)=0.dim_{H}(E_{1,d})=0.

In this paper, we generalize the above results by showing the following.

Theorem 2.

Let m,dm,d\in\mathbb{N} with 1<d1<d and 1md1\leq m\leq d. Then,

dimH(Em,d)=logmlogd.dim_{H}(E_{m,d})=\frac{\log m}{\log d}.

This paper is organized in the following way:

In Section 2, we get a lower bound on the Hausdorff Dimension of Em,dE_{m,d} by calculating the Hausdorff Dimension of a subset of Em,dE_{m,d}.

In Section 3, we prove the upper bound by imitating the proof of Theorem 1 in paper  [mcmullen2010dynamics] using combinatorial arguments, and conclude by proving Theorem 2.

2. Lower Bound

In this section, we define and study two useful invariants of a cycle called crossing number and Digit Portrait, which are directly related to the degree of the cycle. Then we use these properties to construct a subset of Em,dE_{m,d} which has Hausdorff Dimension logmlogd\frac{\log m}{\log d}.

Definition 6 (Crossing).

Let dd be a positive integer greater than 1. Let C={c1,c2,,cn}C=\{c_{1},c_{2},...,c_{n}\} be a cycle for dd-map such that

0<c1<c2<.<cn<10<c_{1}<c_{2}<....<c_{n}<1

Let cn+1=c1c_{n+1}=c_{1}. For any 1in1\leq i\leq n, the pair (ci,ci+1)(c_{i},c_{i+1}) is called a crossing generated by 𝐂\mathbf{C} (or simply a crossing) iff

(5) 0<dci+1(mod 1)<dci(mod 1)<10<dc_{i+1}\text{(mod 1)}<dc_{i}\text{(mod 1)}<1

The total number of such crossings is called the crossing number of C.

(6) η(C)=# (crossings generated by C)\eta(C)=\text{\# (crossings generated by C)}
Remark 3.

If n=1n=1, then there are no crossings. The crossing number is 0 which is the degree of 1-element cycles.

Remark 4.

As we jump on S1S^{1} in the counterclockwise direction and trace the points of the cycle from c1c2.cnc_{1}\rightarrow c_{2}\rightarrow....\rightarrow c_{n} and (jump over 0) back to c1c_{1}, we complete one full circle. When we trace the images of these points, dc1(mod 1)dc2(mod 1)dcn(mod 1)dc_{1}\text{(mod 1)}\rightarrow dc_{2}\text{(mod 1)}\rightarrow......\rightarrow dc_{n}\text{(mod 1)} and back to dc1(mod 1)dc_{1}\text{(mod 1)}, we may jump over 0 multiple times. Each time we jump over 0, we have a crossing. The crossing number of the cycle is equal to its degree. We provide a rigorous proof of this intuitive result below.

Lemma 1.

Let dd be a positive integer greater than 1. Let C={c1,c2,,cn}C=\{c_{1},c_{2},...,c_{n}\} be a cycle for dd-map with 0<c1<c2<.<cn<10<c_{1}<c_{2}<....<c_{n}<1. Then, the crossing number of CC is equal to its degree or

η(C)=deg(C)\eta(C)=deg(C)
Proof.

The case n=1n=1 is obvious, so assume n>1n>1 and let cn+k=ckc_{n+k}=c_{k}, for all k.k\in\mathbb{N}.

Let 1i1<i2<<iη(C)n1\leq i_{1}<i_{2}<...<i_{\eta(C)}\leq n be such that for all 1tη(C)1\leq t\leq\eta(C), the pair (cit,cit+1)(c_{i_{t}},c_{i_{t}+1}) is a crossing generated by CC.

First, we prove deg(C)η(C)deg(C)\geq\eta(C). Assume for the sake of argument that η(C)>deg(C)\eta(C)>deg(C). We can divide S1S^{1} or [0,1)[0,1) in intervals I1,I2,,Ideg(C)I_{1},I_{2},...,I_{deg(C)} such that on each IrI_{r}, there exists a continuous non-decreasing map grg_{r} to [0,1)[0,1) which agrees with the dd-map on IrCI_{r}\cap C.

Note that there are exactly η(C)\eta(C) sets of the type {cit,cit+1,cit+1}.\{c_{i_{t}},c_{i_{t}+1},...c_{i_{t+1}}\}. By our assumption that η(C)>deg(C),\eta(C)>deg(C), we can find a tt such that {cit,cit+1,cit+1}Ir\{c_{i_{t}},c_{i_{t}+1},...c_{i_{t+1}}\}\subset I_{r} for some rr. In other words,

0<gr(cit+1)=dcit+1(mod 1)<dcit(mod 1)=gr(cit)<1,0<g_{r}(c_{i_{t}+1})=dc_{i_{t}+1}\text{(mod 1)}<dc_{i_{t}}\text{(mod 1)}=g_{r}(c_{i_{t}})<1,

a contradiction of the non-decreasing nature of grg_{r}. Thus, deg(C)η(C)deg(C)\geq\eta(C).

Now we prove that deg(C)η(C)deg(C)\leq\eta(C) by constructing an appropriate map f:S1S1f:S^{1}\rightarrow S^{1} of degree η(C)\eta(C). To start, we divide S1S^{1} into η(C)\eta(C) intervals of the type [cit+cit+12,cit+1+cit+1+12]\left[\frac{c_{i_{t}}+c_{i_{t}+1}}{2},\frac{c_{i_{t+1}}+c_{{i_{t+1}}+1}}{2}\right], each of which is mapped onto S1.S^{1}. We take the endpoints of each such interval to 0, and points of CC to their images under the dd-map. In between, we make ff linear. We define

f(x)={dcit+1(mod 1)cit+1cit2(xcit+cit+12)if x[cit+cit+12,cit+1]dci(mod 1)+dci+1(mod 1)dci(mod 1)ci+1ci(xci)if x[ci,ci+1][cit+1,cit+1]dcit+1(mod 1)+1dcit+1(mod 1)cit+1+1cit+12(xcit+1)if x[cit+1,cit+1+cit+1+12]f(x)=\left\{\begin{array}[]{l l}\frac{dc_{i_{t}+1}\text{(mod 1)}}{\frac{c_{i_{t}+1}-c_{i_{t}}}{2}}\ (x-\frac{c_{i_{t}}+c_{i_{t}+1}}{2})&\quad\text{if $x\in\left[\frac{c_{i_{t}}+c_{i_{t}+1}}{2},c_{i_{t}+1}\right]$}\\ &\\ dc_{i}\text{(mod 1)}+{\frac{dc_{i+1}\text{(mod 1)}-dc_{i}\text{(mod 1)}}{c_{i+1}-c_{i}}}\ (x-c_{i})&\quad\text{if $x\in\left[c_{i},c_{i+1}\right]\subset\left[c_{i_{t}+1},c_{i_{t+1}}\right]$}\\ &\\ dc_{i_{t+1}}\text{(mod 1)}+{\frac{1-dc_{i_{t+1}}\text{(mod 1)}}{\frac{c_{i_{t+1}+1}-c_{i_{t+1}}}{2}}}\ (x-c_{i_{t+1}})&\quad\text{if }x\in\left[c_{i_{t+1}},\frac{c_{i_{t+1}}+c_{{i_{t+1}}+1}}{2}\right]\end{array}\right.

where intervals are taken in counter-clockwise direction.

So, ff is a continuous function for all tt, and the restriction of ff gives a bijection

[cit+cit+12,cit+1+cit+1+12)S1.\left[\frac{c_{i_{t}}+c_{i_{t}+1}}{2},\frac{c_{i_{t+1}}+c_{{i_{t+1}}+1}}{2}\right)\leftrightarrow S^{1}.

Since ff has degree η(C)\eta(C) and f|Cf|_{C} agrees with the dd-map, by Definition 3, deg(C)η(C).deg(C)\leq\eta(C). This, combined with above, completes the proof.

Now that we have established the relation between the degree and the crossing number of a cycle, we need a tool to estimate the crossing number. We observe that the crossing number of a cycle is related to the order of points in the cycle, and hence the digits in the base-dd expansion of points of the cycle. We define an invariant of the cycle called Digit Portrait which characterizes these digits.

Definition 7 (Digit Portrait).

Let dd be a positive integer greater than 1. Let CC be a cycle for dd-map. The Digit Portrait of CC is the non-decreasing map F:{0,1,2,,(d1)}{0,1,2,,|C|}F:\{0,1,2,...,(d-1)\}\rightarrow\{0,1,2,...,|C|\} which satisfies

F(j)=|C[0,(j+1)/d)| 0j(d1)F(j)=|\ C\ \cap\ [0,(j+1)/d)\ |\ \ \forall\ 0\leq j\leq(d-1)

or

F(j)=# (elements of C whose base-d expansion starts with a digit less than j+1).F(j)=\text{\# (elements of $C$ whose base-$d$ expansion starts with a digit less than $j+1$)}.

Let dig(C)dig(C) be the number of distinct positive values taken by FF. Note that if a digit jj is absent in the base-dd expansion of a point in CC, then F(j)=0F(j)=0 or F(j)=F(j1)F(j)=F(j-1). So, dig(C)dig(C) is also the number of distinct digits which appear in the base-dd expansion of any point in CC. To estimate the crossing number of CC, we need the second interpretation of dig(C)dig(C).

Example 3.

Consider d=4d=4 and the cycle C={(0.0012¯)4,(0.0120¯)4,(0.1200¯)4,(0.2001¯)4}C=\{(0.\overline{0012})_{4},(0.\overline{0120})_{4},(0.\overline{1200})_{4},(0.\overline{2001})_{4}\}. Note that

0=(0.0)4<(0.0012¯)4<(0.0120¯)4<14=(0.1)4<(0.1200¯)4<24=(0.2)4<(0.2001¯)3<34=(0.3)40=(0.0)_{4}<(0.\overline{0012})_{4}<(0.\overline{0120})_{4}<\frac{1}{4}=(0.1)_{4}<(0.\overline{1200})_{4}<\frac{2}{4}=(0.2)_{4}<(0.\overline{2001})_{3}<\frac{3}{4}=(0.3)_{4}

The Digit Portrait of CC is the map F:{0,1,2,3}{0,1,2,3,4}F:\{0,1,2,3\}\rightarrow\{0,1,2,3,4\} given by:

F(0)=2,F(1)=3,F(2)=4,F(3)=4.F(0)=2,F(1)=3,F(2)=4,F(3)=4.

FF takes the values 2, 3 and 4. So, dig(C)dig(C) is 3. There are exactly 3 digits (0, 1 and 2) which appear in the base-4 expansion of the points in CC.

Now we establish the relation between dig(C)dig(C) and the crossing number of CC.

Lemma 2.

Let dd be a positive integer greater than 1, and C={c1,c2,,cn}C=\{c_{1},c_{2},...,c_{n}\} be a cycle for the dd-map with 0<c1<c2<.<cn<10<c_{1}<c_{2}<....<c_{n}<1 and n>1n>1. Then, the crossing number of CC is at most the number of distinct digits which appear in the base-dd expansion of a point in CC. In other words,

η(C)dig(C).\eta(C)\leq dig(C).
Proof.

Let 1i<n1\leq i<n. Let ci[j1/d,(j1+1)/d)c_{i}\in[j_{1}/d,(j_{1}+1)/d) and ci+1[j2/d,(j2+1)/d)c_{i+1}\in[j_{2}/d,(j_{2}+1)/d). In other words, the base-dd expansions of cic_{i} and ci+1c_{i+1} begin with the digits j1j_{1} and j2,j_{2}, respectively. Note that j1j2j_{1}\leq j_{2} because ci<ci+1c_{i}<c_{i+1}.

If j1=j2=jj_{1}=j_{2}=j, then

0<dci(mod 1)=dcij<dci+1j=dci+1(mod 1)<1.0<dc_{i}\text{(mod 1)}=dc_{i}-j<dc_{i+1}-j=dc_{i+1}\text{(mod 1)}<1.

In this case, (ci,ci+1)(c_{i},c_{i+1}) cannot be a crossing. So, (ci,ci+1)(c_{i},c_{i+1}) is a crossing only if j1<j2,j_{1}<j_{2}, i.e., cic_{i} is the largest element of C[j1/d,(j1+1)/d)C\cap\ [j_{1}/d,(j_{1}+1)/d) and ci+1c_{i+1} is the smallest element of C[j2/d,(j2+1)/d)C\ \cap\ [j_{2}/d,(j_{2}+1)/d).

Thus, there are at most (dig(C)1)i(dig(C)-1)\ i’s for which 1i<n1\leq i<n and (ci,ci+1)(c_{i},c_{i+1}) is a crossing. For some cycles, (cn,c1)(c_{n},c_{1}) is an additional crossing. So, there are at most dig(C)idig(C)\ i’s for which 1in1\leq i\leq n and (ci,ci+1)(c_{i},c_{i+1}) is a crossing. ∎

Together, Lemma 1 and Lemma 2 give a way of estimating the degree of a cycle by looking at the digits in the base-dd expansion of a point in the cycle. Now we use this to get a sufficient condition for a point to be in the closure of the cycles of fixed degree.

Lemma 3.

Let m,dm,d\in\mathbb{N} with 1md1\leq m\leq d and 1<d1<d. Then, any point in S1S^{1} whose base-dd expansion contains at most mm distinct digits lies in Em,dE_{m,d}.

Proof.

Let αS1.\alpha\in S^{1}. Let α=(0.α1α2α3..)d\alpha={(0.\alpha_{1}\alpha_{2}\alpha_{3}.....)}_{d} such that r,αr{b1,b2,,bm}{0,1,,(d1)}.\forall\ r,\alpha_{r}\in\{b_{1},b_{2},...,b_{m}\}\subset\{0,1,...,(d-1)\}. Here, b1,b2,,bmb_{1},b_{2},...,b_{m} are fixed digits in base-dd such that b1<b2<<bm.b_{1}<b_{2}<...<b_{m}.

To prove that α\alpha lies in Em,dE_{m,d}, we will show that for all q,q\in\mathbb{N}, there exists a degree-mm cycle CC for the dd-map that intersects dqd^{-q} neighborhood of α\alpha. Any periodic point whose base-dd expansion contains exactly mm distinct digits is in a cycle of degree at most mm. To get the maximum possible degree, we need maximum possible crossings. This can be achieved with the following construction:

Let NN\in\mathbb{N} such that for all 1tm,1\leq t\leq m, NN is greater than the number of times btb_{t} appears in the first qq digits of α\alpha. Let bt\langle b_{t}\rangle denote btbtbtbt(N times)b_{t}b_{t}b_{t}...b_{t}\ \ (N\text{\ times}). Consider the following point:

(7) c=(0.α1α2α3αqbmb1bm1b1.b2b1bm¯)dc={(0.\overline{\alpha_{1}\alpha_{2}\alpha_{3}...\alpha_{q}\langle b_{m}\rangle\langle b_{1}\rangle\langle b_{m-1}\rangle\langle b_{1}\rangle....\langle b_{2}\rangle\langle b_{1}\rangle b_{m}})}_{d}

Clearly, cc is in dqd^{-q} neighborhood of α\alpha. It is a point of cycle C={c1,c2,cq+2N(m1)+1}C=\{c_{1},c_{2},...c_{q+2N(m-1)+1}\} for the dd-map with 0<c1<c2<<cq+2N(m1)+10<c_{1}<c_{2}<...<c_{q+2N(m-1)+1}.

Now we need to prove that deg(C)=mdeg(C)=m. For each 1tm1\leq t\leq m, let iti_{t} be such that the largest element of CC whose base-dd expansion starts with the digit btb_{t} is citc_{i_{t}}. Note that ci1c_{i_{1}} is at least (0.b1bm)d{(0.b_{1}b_{m})}_{d}. For t>1t>1, cit\ c_{i_{t}} is at least (0.bt)d{(0.\langle b_{t}\rangle)}_{d}. For each 1t<m,cit+11\leq t<m,\ c_{i_{t}+1} is the smallest element of CC whose base-dd expansion starts with the digit bt+1b_{t+1}. Note that the base-dd expansion of cit+1c_{i_{t}+1} starts with 0.btb10.b_{t}\langle b_{1}\rangle.

So, for all 1t<m1\leq t<m,

0<dcit+1(mod 1)<dcit(mod 1)<1,0<dc_{i_{t}+1}\text{(mod 1)}<dc_{i_{t}}\text{(mod 1)}<1,

i.e., (cit,cit+1)(c_{i_{t}},c_{i_{t}+1}) is a crossing.

Note that cq+2N(m1)+1=cimc_{q+2N(m-1)+1}=c_{i_{m}} and cim+1=c1=ci1c_{i_{m}+1}=c_{1}=c_{i_{1}}. So,

0<dcim+1(mod 1)<(0.b2)d<dcim(mod 1)<1,0<dc_{i_{m}+1}\text{(mod 1)}<{(0.b_{2})}_{d}<dc_{i_{m}}\text{(mod 1)}<1,

i.e., (cim,cim+1)(c_{i_{m}},c_{i_{m}+1}) is a crossing.

Thus, the crossing number of CC is at least mm.

η(C)m\eta(C)\geq m

From Lemma 2, we know that the crossing number of CC is at most the number of distinct digits in the base-dd expansion of a point CC, which in this case, is mm.

η(C)dig(C)=m\eta(C)\leq dig(C)=m

Therefore, deg(C)=η(C)=m\deg(C)=\eta(C)=m.

The following result follows immediately.

Proposition 1.

Ed,d=S1E_{d,d}=S^{1}. Thus, dimH(Ed,d)=1=logdlogd.dim_{H}(E_{d,d})=1=\frac{\log d}{\log d}.

Proof.

Note that since base-dd has only dd digits, the set of all points of S1S^{1} whose base-dd expansion contains at most dd distinct digits is S1S^{1} itself. So, we have S1Ed,dS1S^{1}\subset E_{d,d}\subset S^{1} or Ed,d=S1.E_{d,d}=S^{1}.

Now, we use Lemma 3 to construct a subset of Em,dE_{m,d}. Let mm and dd be positive integers with d>1d>1 and 1md1\leq m\leq d, and let Am,dA_{m,d} be the set of points in S1S^{1} whose base-dd expansion contains the digits from 0 to (m1)(m-1) only. Clearly, Am,dEm,dA_{m,d}\subset E_{m,d}.

The structure of Am,dA_{m,d} is similar to the structure of Cantor’s set. Here, we start with X0=[0,1]X_{0}=[0,1]. Write X0X_{0} in dd closed intervals of length 1/d1/d. Let X1X_{1} be union of first mm of these intervals.

X1=i=0m1[i/d,(i+1)/d]X_{1}=\bigcup_{i=0}^{m-1}{[i/d,{(i+1)}/d]}

X1X_{1} is the union of mm intervals of length 1/d1/d. Divide each such interval in dd equal parts and take the first mm in X2.X2X_{2}.\ X_{2} is the union of m2m^{2} intervals of length 1/d21/{d^{2}}.

Repeat the process. If XiX_{i} is the union of mim^{i} intervals of length did^{-i}, then divide each such interval in dd equal parts and take the first mm in Xi+1X_{i+1}.

Am,d=i=0XiA_{m,d}=\bigcap_{i=0}^{\infty}{X_{i}}

Now we calculate the Hausdorff Dimension of Am,dA_{m,d}.

Lemma 4.

dimH(Am,d)=logmlogddim_{H}(A_{m,d})=\frac{\log m}{\log d}.

Proof.

First, we prove that the Hausdorff dimension of Am,dA_{m,d} is at most logmlogd\frac{\log m}{\log d}. Let β>0\beta>0. Note that for all i,Am,dXii,A_{m,d}\subset X_{i}. So, for each ii, we have mim^{i} intervals of length did^{-i} that form a covering of Am,dA_{m,d}. Letting β=logmlogd+ε\beta=\frac{\log m}{\log d}+\varepsilon for some ε>0\varepsilon>0,

limimi(di)β=limimi(diε)((dlogmlogd)i)=limidiε=0\lim_{i\to\infty}{m^{i}{(d^{-i})}^{\beta}}=\lim_{i\to\infty}{m^{i}(d^{-i\varepsilon})((d^{\frac{\log m}{\log d}})^{-i})}=\lim_{i\to\infty}{d^{-i\varepsilon}}=0

This means that, for any β>logmlogd\beta>\frac{\log m}{\log d}, we can cover Am,dA_{m,d} such that the summation of β\beta powers of the lengths of the intervals in the cover is as small as we like. So,

dimH(Am,d)logmlogd.dim_{H}(A_{m,d})\leq\frac{\log m}{\log d}.

Now we need to show that the Hausdorff dimension of Am,dA_{m,d} is at least logmlogd\frac{\log m}{\log d}. Note that we can consider Am,dA_{m,d} as a subset of [0,m/d],[0,m/d], a compact set. So, for any countable cover (Ur)(U_{r}) of Am,dA_{m,d}, we can find finitely many open sets V1,V2,,VpV_{1},V_{2},...,V_{p} such that

r=0Urr=0pVr and\bigcup_{r=0}^{\infty}{U_{r}}\subset\bigcup_{r=0}^{p}{V_{r}}\text{\quad and}
r=0p|Vr|βr=0|Ur|β.\sum_{r=0}^{p}{|V_{r}|^{\beta}}\leq\sum_{r=0}^{\infty}{|U_{r}|^{\beta}}.

We next get a lower bound on r=0p|Vr|β\sum_{r=0}^{p}{|V_{r}|^{\beta}}. Let bb\in\mathbb{N} such that for all r,db|Vr|r,d^{-b}\leq|V_{r}|. Additionally, for all 1ib1\leq i\leq b, let NiN_{i} be the number of VrV_{r}’s which satisfy di|Vr|<di+1d^{-i}\leq|V_{r}|<d^{-i+1}. Observe that if |Vr|<di+1|V_{r}|<d^{-i+1}, then VrV_{r} can intersect at most two intervals in Xi1X_{i-1}. Hence, VrV_{r} can contain at most 2mbi+12m^{b-i+1} intervals in XbX_{b}, which has mbm^{b} intervals. So,

i=0b2mbi+1Nimb\sum_{i=0}^{b}{2m^{b-i+1}N_{i}}\geq m^{b}

or

i=0bmiNi12m\sum_{i=0}^{b}{m^{-i}N_{i}}\geq\frac{1}{2m}

For β=logmlogd\beta=\frac{\log m}{\log d},

r=0p|Vr|βi=0b(di)βNi=i=0bmiNi12m.\sum_{r=0}^{p}{|V_{r}|^{\beta}}\geq\sum_{i=0}^{b}{{(d^{-i})}^{\beta}N_{i}}=\sum_{i=0}^{b}{m^{-i}N_{i}}\geq\frac{1}{2m}.

In other words, the summation of logmlogd\frac{\log m}{\log d} powers of the lengths of the intervals which form a cover of Am,dA_{m,d} has a positive lower bound. Thus,

dimH(Am,d)logmlogd.dim_{H}(A_{m,d})\geq\frac{\log m}{\log d}.

Since Am,dEm,d,A_{m,d}\subset E_{m,d}, we immediately get a lower bound on the Hausdorff Dimension of Em,d:E_{m,d}:

Theorem 3.

Let m,dm,d\in\mathbb{N} with 1<d1<d and 1md1\leq m\leq d. Then,

dimH(Em,d)logmlogddim_{H}(E_{m,d})\geq\frac{\log m}{\log d}

3. Upper Bound

In this section, we first find an upper bound on the number of degree-mm cycles for dd-map which have nn elements. We extend this result to precycles. Then, we find an appropriate covering of Em,dE_{m,d} to prove that its Hausdorff Dimension is at most logmlogd\frac{\log m}{\log d}.

Definition 8 (Partition generated by a cycle).

Let C={c1,c2,,cn}C=\{c_{1},c_{2},...,c_{n}\} be a degree-mm cycle for dd-map such that 0<c1<c2<.<cn<10<c_{1}<c_{2}<....<c_{n}<1. Let σ\sigma be the map on {1,2,.,n}\{1,2,....,n\} which satisfies:

dcr(mod 1)=cσ(r), 1rndc_{r}\text{(mod 1)}=c_{\sigma(r)},\forall\ 1\leq r\leq n

Note that σ\sigma is a permutation of {1,2,.,n}\{1,2,....,n\} because CC is a cycle.

Let 1i1<i2<<imn1\leq i_{1}<i_{2}<...<i_{m}\leq n be such that for all 1tm,(cit,cit+1)1\leq t\leq m,(c_{i_{t}},c_{i_{t}+1}) is a crossing generated by CC. From the ordering of the elements of CC and the definition of crossing, we conclude that:

σ(it)>σ(it+1) and σ(it+1)<σ(it+2)<<σ(it+1), 1t<m\sigma(i_{t})>\sigma(i_{t}+1)\text{ and }\sigma(i_{t}+1)<\sigma(i_{t}+2)<...<\sigma(i_{t+1}),\ \forall\ 1\leq t<m

and

σ(im)>σ(im+1) and σ(im+1)<<σ(n)<σ(1)<<σ(i1)\sigma(i_{m})>\sigma(i_{m}+1)\text{ and }\sigma(i_{m}+1)<...<\sigma(n)<\sigma(1)<...<\sigma(i_{1})

Now we construct a partition of {1,2,..,n}\{1,2,..,n\} using the above property of σ\sigma.

Pt={{σ(r)|it<rit+1}if 1t<m{σ(r)|im<rn}{σ(r)|1ri1}if t=mP_{t}=\left\{\begin{array}[]{l l}\{\sigma(r)|i_{t}<r\leq i_{t+1}\}&\quad\text{if $1\leq t<m$}\\ \{\sigma(r)|i_{m}<r\leq n\}\cup\{\sigma(r)|1\leq r\leq i_{1}\}&\quad\text{if $t=m$}\end{array}\right.

{Pt|1tm}\{P_{t}|1\leq t\leq m\} is a partition of {1,2,..,n}\{1,2,..,n\}, called as the partition generated by CC and it is denoted by P(C)P(C).

Remark 5.

Both P(C)P(C) and the map σ\sigma are useful counting nn-element degree-mm cycles, as we show below.

Example 4.

Let us consider the instance where d=3d=3 and

C={(0.00102¯)3,(0.01020¯)3,(0.02001¯)3,(0.10200¯)3,(0.20010¯)3}.C=\{(0.\overline{00102})_{3},(0.\overline{01020})_{3},(0.\overline{02001})_{3},(0.\overline{10200})_{3},(0.\overline{20010})_{3}\}.

Note that here, n=5n=5. Under the 3-map,

c1\displaystyle c_{1} =(0.00102¯)3(0.01020¯)3=c2\displaystyle=(0.\overline{00102})_{3}\mapsto(0.\overline{01020})_{3}=c_{2}
c2\displaystyle c_{2} =(0.01020¯)3(0.10200¯)3=c4\displaystyle=(0.\overline{01020})_{3}\mapsto(0.\overline{10200})_{3}=c_{4}
c3\displaystyle c_{3} =(0.02001¯)3(0.20010¯)3=c5\displaystyle=(0.\overline{02001})_{3}\mapsto(0.\overline{20010})_{3}=c_{5}
c4\displaystyle c_{4} =(0.10200¯)3(0.02001¯)3=c3\displaystyle=(0.\overline{10200})_{3}\mapsto(0.\overline{02001})_{3}=c_{3}
c5\displaystyle c_{5} =(0.20010¯)3(0.00102¯)3=c1\displaystyle=(0.\overline{20010})_{3}\mapsto(0.\overline{00102})_{3}=c_{1}

So, σ\sigma is the permutation of {1,2,3,4,5}\{1,2,3,4,5\} which takes 1, 2, 3, 4, 5 to 2, 4, 5, 3, 1 respectively. (c3,c4)(c_{3},c_{4}) and (c4,c5)(c_{4},c_{5}) are the crossings generated by the cycle. So, i1=3i_{1}=3 and i2=4i_{2}=4, and the partition generated by CC is given by:

P(C)={P1,P2} where P1={σ(r)|3<r4},P2={σ(r)|4<r5}{σ(r)|1r3}P(C)=\{P_{1},P_{2}\}\text{ where }P_{1}=\{\sigma(r)|3<r\leq 4\},P_{2}=\{\sigma(r)|4<r\leq 5\}\cup\{\sigma(r)|1\leq r\leq 3\}

or

P(C)={{3},{1,2,4,5}}P(C)=\{\{3\},\{1,2,4,5\}\}

Now, we will show that if some properties of a cycle such as degree, the partition it generates and its digit portrait are given, then we can construct the cycle (find its points). Later, we will use this result to count the number of cycles of fixed degree and cardinality.

Lemma 5.

Given positive integers d,m,nd,m,n, P={P1,P2,,Pm}P=\{P_{1},P_{2},...,P_{m}\} which is a partition of {1,2,,n}\{1,2,...,n\}, positive integer i1<|Pm|i_{1}<|P_{m}| and a non-decreasing map F:{0,1,2,,(d1)}{0,1,2,,n}F:\{0,1,2,...,(d-1)\}\rightarrow\{0,1,2,...,n\} with F(d1)=nF(d-1)=n, there exists at most one cycle CC for dd-map such that:

  1. (1)

    C={c1,c2,,cn}C=\{c_{1},c_{2},...,c_{n}\} with 0<c1<c2<.<cn<10<c_{1}<c_{2}<....<c_{n}<1

  2. (2)

    deg(C)=mdeg(C)=m

  3. (3)

    P(C)=PP(C)=P or PP is the partition generated by CC

  4. (4)

    If, for all 1<tm1<t\leq m, it=it1+|Pt|i_{t}=i_{t-1}+|P_{t}|, then (ci,ci+1)(c_{i},c_{i+1}) is a crossing for all i{i1,i2,,im}i\in\{i_{1},i_{2},...,i_{m}\}.

  5. (5)

    FF is the Digit Portrait of CC.

Proof.

Suppose CC is a cycle for dd-map which satisfies all the conditions above, and let σ\sigma be the map on {1,2,.,n}\{1,2,....,n\} given by:

σ(r)={(rit)th element of Ptif it<rit+1 and 1t<m(rim)th element of Pmif im<r(r+nim)th element of Pmif r<i1\sigma(r)=\left\{\begin{array}[]{l l}{(r-i_{t})}^{\text{th}}\text{ element of }P_{t}&\quad\text{if $i_{t}<r\leq i_{t+1}$ and $1\leq t<m$}\\ {(r-i_{m})}^{\text{th}}\text{ element of }P_{m}&\quad\text{if $i_{m}<r$}\\ {(r+n-i_{m})}^{\text{th}}\text{ element of }P_{m}&\quad\text{if $r<i_{1}$}\\ \end{array}\right.

where, for all 1tm1\leq t\leq m, the elements of PtP_{t} are in increasing order. Note that σ\sigma is uniquely determined by m,n,Pm,n,P and i1i_{1}. Comparing this with the definition of partition generated by a cycle, we conclude that

dcr(mod 1)=cσ(r), 1rndc_{r}\text{(mod 1)}=c_{\sigma(r)},\forall\ 1\leq r\leq n

Let b:{1,2,.,n}{0,1,2,.,(d1)}b:\{1,2,....,n\}\rightarrow\{0,1,2,....,(d-1)\} given by:

b(r)={0if r<F(0)jif F(j1)<rF(j) and 1j(d1)b(r)=\left\{\begin{array}[]{l l}0&\quad\text{if $r<F(0)$}\\ j&\quad\text{if $F(j-1)<r\leq F(j)$ and $1\leq j\leq(d-1)$}\end{array}\right.

Note that bb is uniquely determined by n,dn,d and kk. Comparing this with the definition of Digit Portrait, we conclude that

cr[b(r)d,b(r)+1d), 1rnc_{r}\in[\ \frac{b(r)}{d},\frac{b(r)+1}{d}\ ),\ \forall\ 1\leq r\leq n

or

b(r)=the first digit in the base-d expansion of cr,for all 1rnb(r)=\text{the first digit in the base-}d\text{ expansion of }c_{r},\text{for all}\ 1\leq r\leq n

Also,

b(σ(r))=the first digit in the base-d expansion of cσ(r)b(\sigma(r))=\text{the first digit in the base-}d\text{ expansion of }c_{\sigma(r)}
= the second digit in the base-d expansion of cr\quad\quad=\text{ the second digit in the base-}d\text{ expansion of }c_{r}

Thus, we conclude that

cr=(0.b(r)b(σ(r))b(σ2(r))..b(σd1(r))¯)dc_{r}={(0.\overline{b(r)b(\sigma(r))b(\sigma^{2}(r)).....b(\sigma^{d-1}(r))}\ )}_{d}

In other words, the cycle CC is uniquely determined by σ\sigma and bb.

Remark 6.

(ci,ci+1)(c_{i},c_{i+1}) can be a crossing only if the first digits of the base-dd expansions of cic_{i} and ci+1c_{i+1} differ. So, cycle CC satisfying all conditions in the above lemma can exist only if

{F(0),F(1),,F(d1)}{i1,i2,im}{n}\{F(0),F(1),...,F(d-1)\}\subset\{i_{1},i_{2},...i_{m}\}\cup\{n\}

We will use this in the proof of the following lemma.

Lemma 6.

Let d,m,nd,m,n\in\mathbb{N} with 1md1\leq m\leq d and n,d>1n,d>1. Then the number of cycles CC for dd-map satisfying |C|=n|C|=n and deg(C)=mdeg(C)=m is at most O(ndm+1mn1).O(n^{d-m+1}m^{n-1}).

Proof.

d,m,nd,m,n are given. Now we need P={P1,P2,,Pm}P=\{P_{1},P_{2},...,P_{m}\} which is a partition of {1,2,,n}\{1,2,...,n\}, positive integer i1<|Pm|i_{1}<|P_{m}| and a non-decreasing map F:{0,1,2,,(d1)}{0,1,2,,n}F:\{0,1,2,...,(d-1)\}\rightarrow\{0,1,2,...,n\} with F(d1)=nF(d-1)=n to fix a degree-mm, nn-element cycle.

Let TT be the set of ordered (m+1)(m+1)-tuples (P1,P2,,Pm,i1)(P_{1},P_{2},...,P_{m},i_{1}) such that P={P1,P2,,Pm}P=\{P_{1},P_{2},...,P_{m}\} is a partition of {1,2,,n}\{1,2,...,n\}, i1i_{1}\in\mathbb{N} and i1|Pm|i_{1}\leq|P_{m}|. We want an upper bound on the size of T, denoted here as #(T)\#(T).

#(T)\displaystyle\#(T) =a1+a2++am=naiam(na1)(na1a2)(na1a2am1am)\displaystyle=\sum_{\begin{subarray}{c}a_{1}+a_{2}+...+a_{m}=n\\ a_{i}\in\mathbb{N}\end{subarray}}{a_{m}\ {n\choose{a_{1}}}{{n-a_{1}}\choose{a_{2}}}...{{{n-a_{1}-a_{2}-...-a_{m-1}}\choose{a_{m}}}}}
=a1+a2++am=naiamn!a1!a2!.am!\displaystyle=\sum_{\begin{subarray}{c}a_{1}+a_{2}+...+a_{m}=n\\ a_{i}\in\mathbb{N}\end{subarray}}{a_{m}\cdot\frac{n!}{a_{1}!\ a_{2}!\ ....a_{m}!}}
=a1+a2++am=nain!a1!a2!.am1!(am1)!\displaystyle=\sum_{\begin{subarray}{c}a_{1}+a_{2}+...+a_{m}=n\\ a_{i}\in\mathbb{N}\end{subarray}}{\frac{n!}{a_{1}!\ a_{2}!\ ....a_{m-1}!\ (a_{m}-1)!}}
a1+a2++am1+(am1)=n1ai{0}n(n1)!a1!a2!.am1!(am1)!\displaystyle\leq\sum_{\begin{subarray}{c}a_{1}+a_{2}+...+a_{m-1}+(a_{m}-1)=n-1\\ a_{i}\in\mathbb{N}\cup\{0\}\end{subarray}}{n\cdot\frac{(n-1)!}{a_{1}!\ a_{2}!\ ....a_{m-1}!\ (a_{m}-1)!}}
=nmn1\displaystyle=n\cdot m^{n-1}

After fixing an element of TT, we need to fix a non-decreasing map FF from {0,1,2,,(d1)}\{0,1,2,...,(d-1)\} to {0,1,2,,n}\{0,1,2,...,n\} with F(d1)=nF(d-1)=n and {F(0),F(1),,F(d1)}{i1,i2,im}{n}\{F(0),F(1),...,F(d-1)\}\subset\{i_{1},i_{2},...i_{m}\}\cup\{n\}. Observe that there are (n+1)(dm1)(n+1)^{(d-m-1)} or (n+1)(dm)(n+1)^{(d-m)} choices for FF, depending on if nn is in {i1,i2,im}\{i_{1},i_{2},...i_{m}\} or not.

Using Lemma 5, we conclude that the number of degree-mm nn-element cycles for dd-map is at most O(ndm+1mn1)O(n^{d-m+1}m^{n-1}).

To conclude this section (and to show our desired result), we first introduce precycles.

Definition 9 (Precycle).

Let dd be a positive integer greater than 1. A finite set CPS1C_{P}\subset S^{1} is called a precycle for dd-map, if and only if CP={cdi (mod 1)|i{0}}C_{P}=\{c\cdot d^{i}\text{ (mod 1)}|i\in\mathbb{N}\cup\{0\}\} for some cS1c\in S^{1}. In other words, a precycle is the forward orbit of a rational point in S1S^{1}. It can be written in terms of base-dd expansion of its points as

CP={(0.brbr+1bn1b1b2.bn2¯)d|1rn1}{(0.bkbk+1bn2b1.bk1¯)d|1kn2},C_{P}=\{(0.b_{r}b_{r+1}...b_{n_{1}}\overline{b_{1}^{\prime}b_{2}^{\prime}....b_{n_{2}}^{\prime}})_{d}|1\leq r\leq n_{1}\}\cup\{(0.\overline{b_{k}^{\prime}b_{k+1}^{\prime}...b_{n_{2}}b_{1}^{\prime}....b_{k-1}^{\prime}})_{d}|1\leq k\leq n_{2}\},

where n1n_{1} is a fixed non-negative integer, n2n_{2} is a fixed positive integer and all br,bkb_{r},b_{k}^{\prime} are fixed digits in {0,1,2,,d1}\{0,1,2,...,d-1\}.

Remark 7.

Every precycle CPC_{P} includes a cycle. If n1=0n_{1}=0, then CPC_{P} itself is a cycle.

Example 5.

Consider d=2d=2 and CP={13,23,56}C_{P}=\{\frac{1}{3},\frac{2}{3},\frac{5}{6}\}. CPC_{P} can be written as

CP={56di(mod 1)|i{0}}C_{P}=\{\frac{5}{6}\cdot d^{i}\text{(mod 1)}|i\in\mathbb{N}\cup\{0\}\}

or

CP={56}{13,23}={(0.110¯)2}{(0.01¯)2,(0.10¯)2}C_{P}=\{\frac{5}{6}\}\cup\{\frac{1}{3},\frac{2}{3}\}=\{(0.1\overline{10})_{2}\}\cup\{(0.\overline{01})_{2},(0.\overline{10})_{2}\}
Remark 8.

We can define degree, crossing, crossing number, digit portrait and partition for a precycle simply by replacing CC by CPC_{P} in the respective definitions.
In the case where CPC_{P} is a precycle but not a cycle, the map σ\sigma in the definition of partition (Definition 8) is not a permutation. Under σ\sigma, one element of {1,2,,n}\{1,2,...,n\} has no preimages, one has two preimages and all other elements have exactly one preimage.

Lemma 7.

Let d,m,nd,m,n\in\mathbb{N} with 1md1\leq m\leq d. Then the number of precycles CPC_{P} for the dd-map satisfying |CP|=n|C_{P}|=n and deg(CP)=mdeg(C_{P})=m is at most O(ndm+3mn1)O(n^{d-m+3}m^{n-1}).

Proof.

Similar to the argument used to prove Lemma 6. Note that the increase in the exponent of nn is due to the small change in the nature of the map σ\sigma in the definition of partition (Definition 8).

Finally, we find a suitable covering of Em,dE_{m,d} and get an upper bound on its Hausdorff Dimension, as desired.

Theorem 4.

Let m,dm,d\in\mathbb{N} with 1<d1<d and 1md1\leq m\leq d. Then

dimH(Em,d)logmlogddim_{H}(E_{m,d})\leq\frac{\log m}{\log d}
Proof.

Fix a positive integer n>1n>1, and let

𝒞P(n)={CP|CP is a precycle for d-map,deg(CP)m,|CP|n}.\mathscr{C}_{P}(n)=\{C_{P}|C_{P}\text{ is a precycle for $d$-map},deg(C_{P})\leq m,|C_{P}|\leq n\}.

From Lemma 7, we have |𝒞P||\mathscr{C}_{P}| is at most O(ndm+4mn)O(n^{d-m+4}m^{n}) (note the change in exponent due to the inequality).

Let CC be a degree-mm cycle for dd-map, and cCc\in C. There exists a degree-mm map f:S1S1f:S^{1}\rightarrow S^{1} which agrees with dd-map on CC. Letting fnf^{n} denote ff composed nn times, we note that |(fn)(c)|=dn|(f^{n})^{\prime}(c)|=d^{n}. See McMullen  [mcmullen2010dynamics] for details. It follows there exists a point xx in O(dn)O(d^{-n}) neighborhood of cc for which two of x,f(x),f2(x),,fn(x)x,f(x),f^{2}(x),...,f^{n}(x) coincide. Generalizing, each point of any degree-mm cycle for the dd-map lies in O(dn)O(d^{-n}) neighborhood of an element in 𝒞P\mathscr{C}_{P}. Thus, Em,dE_{m,d} lies in a O(dn)O(d^{-n}) neighborhood of 𝒞P\mathscr{C}_{P} which has at most O(ndm+4mn)O(n^{d-m+4}m^{n}) elements.

Let β=logmlogd+ε\beta=\frac{\log m}{\log d}+\varepsilon for some ε>0.\varepsilon>0.

limn(dn)βndm+4mn\displaystyle\lim_{n\to\infty}{{(d^{-n})}^{\beta}n^{d-m+4}m^{n}} =limn(dlogmlogd)ndnεndm+4mn\displaystyle=\lim_{n\to\infty}{(d^{\frac{\log m}{\log d}})^{-n}d^{-n\varepsilon}n^{d-m+4}m^{n}}
=limndnεndm+4\displaystyle=\lim_{n\to\infty}{d^{-n\varepsilon}n^{d-m+4}}
=0\displaystyle=0

For any β>logmlogd\beta>\frac{\log m}{\log d}, we can cover Em,dE_{m,d} such that the summation of β\beta powers of the lengths of the intervals in the cover is as small as we like. So, dimH(Em,d)logmlogddim_{H}(E_{m,d})\leq\frac{\log m}{\log d}.

We are now ready to prove Theorem 2.

Theorem 2.

Let m,dm,d\in\mathbb{N} with 1<d1<d and 1md1\leq m\leq d. Then,

dimH(Em,d)=logmlogd.dim_{H}(E_{m,d})=\frac{\log m}{\log d}.
Proof.

Follows from Theorems 3 and 4. ∎

Acknowledgment

The authors would like to thank Prof. Curtis McMullen for suggesting the problems and giving valuable comments. This project was undertaken by M.T. in SPUR (the Summer Program in Undergraduate Research) at MIT Mathematics Department in Summer 2014. M.T. would like to thank Prof. David Jerison and Prof. Pavel Etingof for their guidance and insightful discussions during the program. We also thank our mentor Xuwen Zhu for her help and support for completing this paper.

References