# J. Korean Math. Soc. 31 (1994), No. 4, pp.

539–544

HAUSDORFF DIMENSION OF SOME SPECIFIC PERTURBED CANTOR SET IN SOO BAEK
AND

SANG HUN LEE

1. Introduction We [1] investigated the Hausdorff dimension and the packing dimension of a certain perturbed Cantor set whose ratios are uniformly bounded. In this paper, we consider a speciﬁc perturbed Cantor set whose ratios are not necessarily uniformly bounded but satisfy some other conditions. In fact, in the hypothesis, only the condition of the uniform boundedness of ratios on the set is substituted by a “∗-condition". We use energy theory related to Hausdorff dimension in this study while we [1] used Hausdorff density theorem to ﬁnd the Hausdorff dimension of some perturbed Cantor set. In the end, we give an example which explains aforementioned facts. 2. Preliminaries We recall the deﬁnition of perturbed Cantor set of [1]. Let Iφ =[0,1]. We can obtain the left subinterval Iσ,1 and the right subinterval Iσ,2 of Iσ deleting middle open subinterval of Iσ inductively for each σ ∈ {1, 2}n , where n = 0, 1, 2, · · · . Consider E n = ∪σ ∈{1,2}n Iσ . Then {E n } is a decreasing sequence of closed sets. For each n, we ﬁx | Iσ,1 | / | Iσ |= an+1 and | Iσ,2 | / | Iσ |= bn+1 for all σ ∈ {1, 2}n ,where | I | denotes the diameter of I. We call F = ∞ E n a perturbed Cantor set. n=0 We recall the s-dimensional Hausdorff measure of F : H s (F) = lim Hδs (F),
δ→0

Received June 17, 1993. Revised July 10, 1994. Supported in part by TGRC-KOSEF and the Basic Science Research Institute program, Ministry of Education, Korea, 1994, project NO BSRI-94-1401

540

In Soo Baek and Sang Hun Lee

where Hδs (F) = inf{ ∞ | Un |s : {Un }∞ is a δ-cover of F}, and the n=1 n=1 Hausdorff dimension of F : dim H (F) = sup{s > 0 : H s (F) = ∞} (= inf{s > 0 : H s (F) = 0})(see[2]). We note that if {an } and {bn } are given, then a perturbed Cantor set F is determined, vice versa. We are now ready to study the ratio geometry of the perturbed Cantor set. 3. Main results In this section, F means a perturbed Cantor set determined by {an } and {bn }. We introduce ∗-condition which plays an important role in the energy theory related to Hausdorff dimension. Let s ∈ (0, 1). F is said to satisfy ∗-condition for s, if for each δ > 0 there exist positive < 1 and N such that k (ais + bis ) k [max(ai , i=N i=1 bi )]δ < (1 − )k−N for all k ≥ N . Before going into the main theorem, it is fruitful to know some properties of the ratios {an }, {bn } of a perturbed Cantor set F. LEMMA 1. Let {an }, {bn } be sequences in (0,1) and dn = 1 − (an + bn ) > 0 for each n. Then
s s (1) there exists unique sn such that ann + bnn = 1 for each n = 1, 2, · · · . (2) 0 ≤ s ≤ s ≤ 1, where s = lim infn→∞ sn , s = lim supn→∞ sn . (3) if lim inf n→∞ an > 0 and lim infn→∞ bn > 0, then lim supn→∞ s s (an + bn ) = 1. s s (4) if lim inf n→∞ n log(ak + bk ) is ﬁnite for some s , then s ≤ s ≤ s k=1 . s s (5) if lim inf n→∞ n log(ak + bk ) is ﬁnite for some s , then there is k=1 some B < ∞ such that lim inf n→∞ n log(ais +bis ) < B for any k. i=k s s (6) if lim supn→∞ (an + bn ) ≤ 1 for some s and lim infn→∞ dn > 0, then F satisﬁes the ∗-condition for s .

Proof. (1) - (5) follow immediately from the deﬁnitions of lim inf and lim sup. For (6), we can ﬁnd d > 0 such that an + bn ≤ 1 − d for all n ≥ N1 for some N1 since lim infn→∞ dn > 0. Given δ > 0, we choose

Hausdorff dimension of some speciﬁc perturbed Cantor set

541

0 < < 0 such that (1 − d)δ < 1 − 2 . For such > 0, there is N ≥ s s N1 such that an + bn < 1 + for all n ≥ N . Therefore for all k ≥ N , k (ais + bis ) k [max(ai , bi )]δ < k (ais + bis ) k (1 − d)δ < i=N i=1 i=N i=1 k (1 + )(1 − 2 ) < k (1 + )( 1− ) < (1 − )k−N . i=N i=N 1+
s s THEOREM 2. If lim infn→∞ n log(ak + bk ) is ﬁnite for some s and k=1 lim infn→∞ dn > 0 and F satisﬁes the ∗-condition for s , then dim H (F) = s.

Proof. We deﬁne a set function µ by µ(Iσ ) = lim inf
n→∞ I ∈Iσ ∩E n

|I |s

for each σ ∈ {1, 2}k , where k = 1, 2, · · · . Clearly µ(Iσ ) = µ(Iσ,1 ) + µ(Iσ,2 ) for all σ . Then µ is extended to a mass distribution on F (see [2,Proposition1.7]). For x, y ∈ F, we denote x∧ y by the interval that both x and y belong to a common interval of E n with greatest integer n. Since lim infn→∞ dn > 0, we ﬁnd d > 0 such that ak + bk ≤ 1 − d for all k ≥ N1 for some N1 . Then for σ ∈ {1, 2}k where k ≥ N1 , the gap of Iσ,1 and Iσ,2 is not less than d|Iσ |. Let 0 < t < s. By slight abuse of notation, we write I ∈ E k to mean that the interval I is one of the basic intervals Iσ of E k , where σ ∈ {1, 2}k . If x∧ y ∈ E k and k ≥ N1 , then for σ ∈ {1, 2}k , |x − y|−t dµ(x)dµ(y) ≤2d
−t x∧ y=Iσ |Iσ |−t µ(Iσ,1 )µ(Iσ,2 ) n→∞

=2d −t |Iσ |−t lim inf
I ∈Iσ,1 ∩E n

|I |s lim inf
n→∞ I ∈Iσ,2 ∩E n

|I |s

≤2d −t |Iσ |−t lim inf
n→∞ I ∈Iσ,1 ∩E n ,I ∈Iσ,2 ∩E n n

|I |s |I |s

=2d

−t

lim inf
n→∞ i=k+2 n

(ais + bis )2 |Iσ |−t |Iσ,1 |s |Iσ,2 |s (ais + bis )2 |Iσ |2s−t .
i=k+2

≤2d −t lim inf
n→∞

542

In Soo Baek and Sang Hun Lee

Using Lemma 1 (5), for σ ∈ {1, 2}k where k ≥ N1 , we obtain |x − y|−t dµ(x)dµ(y)
x∧ y=Iσ

≤2d

−t

B|Iσ |2s−t for some B < ∞.

Since F satisﬁes the ∗-condition for s, there exist positive < 1 and N ≥ N1 such that k (ais + bis ) k [max(ai , bi )]s−t < (1 − )k−N for all k ≥ N . i=N i=1 Therefore for all k ≥ N , we have |I |2s−t =
I ∈E k I ∈E k k

|I |s |I |s−t {(ais + bis )[max(ai , bi )]s−t }
i=1

<α(1 − )k−N , where α =
N −1 s i=1 (ai

+ bis ). Thus for k ≥ N , we have |x − y|−t dµ(x)dµ(y)
x∧ y=I ∞

∞ k=N I ∈E k

≤2d −t B

|I |2s−t α
k=N I ∈E k

≤2d −t B . Since dn > 0 for each n, d ∗ = min1≤k≤N dk exists. If x∧ y ∈ E k and k < N , then we similarly obtain |x − y|−t dµ(x)dµ(y)
x∧ y=Iσ ∗ −t ≤2(d ) |Iσ |−t µ(Iσ,1 )µ(Iσ,2 )

≤2(d ∗ )−t B|Iσ |2s−t

Hausdorff dimension of some speciﬁc perturbed Cantor set

543

for σ ∈ {1, 2}k . Thus
N −1 k=0 I ∈E k x∧ y=I N −1

|x − y|−t dµ(x)dµ(y) |I |2s−t
k=0 I ∈E k N −1

≤2(d ) B ≤2(d ) B ≤2(d ) B2 . Hence
∗ −t k=0 N ∗ −t

∗ −t

2k

|x − y|−t dµ(x)dµ(y)
F ∞ F

=
k=0 I ∈E k N −1 x∧ y=I

|x − y|−t dµ(x)dµ(y) |x − y|−t dµ(x)dµ(y)
x∧ y=I

=
k=0 I ∈E k ∞

+
k=N I ∈E k x∧ y=I

|x − y|−t dµ(x)dµ(y) α < ∞.

≤2(d ∗ )−t B2 N + 2d −t B

Since µ is a mass distribution on F, dim H (F) ≥ s (see [2,Theorem 4.13 (a)]). s Now dim H (F) ≤ s follows easily from the fact that lim infn→∞ n log(ak + k=1 s bk ) < ∞ and lim infn→∞ dn > 0 .
s s COROLLARY 3. If lim inf n→∞ n log(ak +bk ) is ﬁnite, lim infn→∞ dn > k=1 s s 0 and lim supn→∞ (an + bn ) ≤ 1, then dim H (F) = s.

Proof. It follows immediately from Theorem 2 and Lemma 1 (6).

544

In Soo Baek and Sang Hun Lee

EXAMPLE 4. Consider a sequence {n k }∞ of natural numbers such that k=1 n k+1 − n k > k + 1 for all k. Let 0 < < 1 and bn k = 1 − for all k. 4 We also take a decreasing sequence {an k }∞ such that 0 < an k ≤ 2 with k=1
2 limk→∞ an k = 0, and log(an2k + bn k ) < 0. Let bn k +i = 1 , 1 ≤ i ≤ k for 4 1 each k. For each k and 1 ≤ i ≤ k , we ﬁnd an k +i such that log{(an k +i ) 2 + 2 (bn k +i ) 2 } = δkk , where δk = − log(an2k + bn k ). We put an = bn = 1 if 4 lim n = n k , n k + 1, · · · , n k + k, where k = 1, 2, · · · . Then √ infn→∞ an = 0, 3−3 and lim infn→∞ dn > 0. In fact, dn ≥ min( 2 , 1 − { 1 + ( 4 6 )2 }) > 0 for all 4 1 2 n. Clearly lim supn→∞ (an2 + bn ) = 1 and lim infn→∞ n log(ak2 + bk2 ) = k=1 1 log(1 − ) 2 . Using Corollary 3, we obtain dim H (F) = 1 . 2 1 1 1 1 1 1 1 1

References
[1] I. S. Baek, Dimensions of the perturbed Cantor set, Real Analysis Exchange 19(1) (19931994), 269-273. [2] K. J. Falconer, Fractal geometry, (John Wiley and Sons. 1990).

SANG HUN LEE Department of Mathematics Kyungpook National University Taegu 702-701, Korea IN SOO BAEK Department of Mathematics Pusan University of Foreign Studies Pusan 608-738, Korea

Sign up to vote on this title