betaFeedback

WorksMathematics

Same Gaps, Different Sets: A Survey on Complimentary sets and their Dimensions

Abstract

Let a={a_n} be a non-increasing sequence of positive real numbers adding up to 11. Cutting out of the unit interval disjoint open gaps of lengths a1,a2,…a_1,a_2,\dots, one for each term, leaves a compact set of Lebesgue measure zero, and since the gaps can be placed in infinitely many ways, there are infinitely many such sets. We call them the complementary sets of aa and write Ca\mathfrak{C}_a for the family of all of them. Every set in Ca\mathfrak{C}_a is compact and totally disconnected, and all of them share the same gap lengths; what distinguishes one from another is only the arrangement of the gaps. Among them, the perfect sets are exactly the Cantor sets with those gaps, and they form a distinguished subclass. The central question of this survey is what such a rearrangement can change and what it cannot. The answer depends sharply on the notion of size: the box dimensions are constant on Ca\mathfrak{C}_a, while the Hausdorff dimension attains every value in a closed interval whose endpoints are realised by two distinguished members, the Cantor set CaC_a built by placing the gaps in a Cantor-like pattern, and the countable decreasing rearrangement DaD_a. We survey how this dichotomy persists, and how the roles of CaC_a and DaD_a are exchanged, across the intermediate dimensions of Falconer, Fraser and Kempton, the Assouad-type and ΦΦ-dimensions, and the Hausdorff and packing hh-measures associated with a general dimension function; we also describe the almost sure behaviour of a randomly rearranged complementary set, where a sharp threshold in the depth of the scales examined separates two incompatible regimes. A final part restricts attention to the Cantor subclass and turns to the dynamical description of the sets C_a.

2020 Mathematics Subject Classification. Primary 28A80; Secondary 28A78, 37C45.

The research of the authors was partially supported by Grants PIP 202287/22 (CONICET), and UBACyT 2022-154 (UBA). Nicolás Angelini was also partially supported by PROICO 3-0720 “Análisis Real y Funcional, Ec.-Diferenciales.

We thank the Fundación Williams for partially supporting our visit to IMASL.

Introduction

Start with a non-increasing sequence a={an}a=\{a_n\} of positive real numbers adding up to 1, and remove from the unit interval [0,1][0,1] disjoint open gaps of lengths a1,a2,…a_1,a_2,\ldots, one for each term of the sequence. Whatever is left is a compact set of Lebesgue measure zero that contains the endpoints 0 and 1. The sequence, however, does not tell us which set we get: the same lengths can be laid out along [0,1][0,1] in infinitely many ways, and each arrangement produces a different set. All these sets together form the family Ca\mathcal{C}_a of complementary sets, or cut-out sets, of aa. Nothing is lost by this point of view: an infinite compact subset of [0,1][0,1] of measure zero containing the endpoints is determined by the open intervals of its complement, so it is a complementary set of exactly one such sequence, the sequence of lengths of those intervals. This family is the object of the survey. Every member of Ca\mathcal{C}_a is compact, totally disconnected and null, but it need not be perfect. The perfect members are exactly the Cantor sets with gap lengths aa; they form a distinguished subclass of Ca\mathcal{C}_a, and we shall restrict to it only when the mathematics requires it. Two members of Ca\mathcal{C}_a are canonical. Placing the gaps in a Cantor-like pattern gives a perfect set CaC_a, the Cantor set associated with aa; placing them in decreasing order from one endpoint gives a countable set DaD_a with a single accumulation point, which is not a Cantor set at all. Both have exactly the same gap lengths, and yet, as soon as one measures them, they could hardly be more different.

The question that organises this survey is therefore the following: given the gap lengths, what is determined for every complementary set in Ca\mathcal{C}_a, and what is left free to the rearrangement? The answer depends entirely on which notion of size one uses, and the resulting picture is surprisingly varied.

Some quantities are rigid. Falconer [17] observed that all sets in Ca\mathcal{C}_a have comparable covering numbers at every scale, so the upper and lower box dimensions are constant on the whole family: no rearrangement, perfect or not, can change them. Others vary as much as they possibly can. Besicovitch and Taylor [10] proved that the Hausdorff dimension of the sets in Ca\mathcal{C}_a sweeps the full interval [0,dim⁡HCa][0,\dim_H C_a], with CaC_a maximal and DaD_a minimal, and that every intermediate value is attained. This contrast—one dimension constant, another taking every admissible value—is the starting point of everything that follows.

Between these two extremes lie the intermediate dimensions of Falconer, Fraser and Kempton, obtained by restricting the diameters of a cover to a geometric window governed by a parameter θ\theta. Angelini and Molter [2] computed them for CaC_a and DaD_a and showed that these two sets again bracket the whole family Ca\mathcal{C}_a, with the full interval attained for every fixed θ\theta: a θ\theta-parametrised version of the Besicovitch–Taylor theorem, degenerating to it at θ=0\theta=0 and to the rigid box dimension at θ=1\theta=1.

Assouad-type dimensions reverse the picture. They record the extreme local behaviour of a set rather than its global size, and for the Assouad dimension and its upper refinements it is CaC_a that minimises and DaD_a that maximises over Ca\mathcal{C}_a, while for the lower Assouad dimension DaD_a is minimal and CaC_a maximal. García, Hare and Mendivil established this for dim⁡A\dim_A and dim⁡L\dim_L [23] and then, through the family of Φ\Phi-dimensions [24], for a whole one-parameter range interpolating towards the box dimensions. The same authors [25] identified a sharp threshold in the depth of the scales examined, at log⁡∣log⁡x∣/∣log⁡x∣\log|\log x|/|\log x|, separating two incompatible almost sure behaviours of a randomly chosen complementary set in Ca\mathcal{C}_a: below it the Assouad dimensions are almost surely extremal regardless of Ca\mathcal{C}_a, above it they almost surely agree with those of Ca\mathcal{C}_a.

A further refinement replaces the exponent ss by a general gauge hh. Here the distinguished Cantor set Ca\mathcal{C}_a comes to the fore. Cabrelli, Mendivil, Molter and Shonkwiler [14] showed that every sequence admits a gauge hah_a making Ca\mathcal{C}_a an hah_a-set, even when no power function does, and García, Molter and Scotto [26] characterised which gauges measure Ca\mathcal{C}_a and exhibited Cantor sets with the same dimension that only a gauge can tell apart. Returning to the whole family, the reversal appears once more at this finer level: Hare, Mendivil and Zuberman [27] proved that, up to multiplicative constants, Ca\mathcal{C}_a has the maximal Hausdorff hh-measure and, simultaneously, the minimal packing hh-premeasure among all complementary sets in Ca\mathcal{C}_a, and that the packing measures attained across the family form either {0}\{0\} or all of [0,∞][0,\infty], with nothing in between; moreover, prescribed values of these measures can be realised by perfect sets, that is, within the Cantor subclass. Not every compact set, however, is measured by some gauge, and [13] shows that this failure is not a rarity.

The recurring theme is that Ca\mathcal{C}_a and Da\mathcal{D}_a are extremal in Ca\mathcal{C}_a, exactly or up to constants, for every notion of size considered here, but which of the two is the maximum depends on whether the notion is global or local, covering-based or packing-based. In most cases where a full description of the attained values is available it takes the same shape: a closed interval with endpoints at those two sets. The packing measures are the notable exception, with the all-or-nothing dichotomy described above.

So far the sequence aa has been treated as a purely combinatorial datum, and the results have concerned the whole family Ca\mathcal{C}_a. The sequence can also be read dynamically, and for this the Cantor subclass, and in fact the single set Ca\mathcal{C}_a, is the natural setting. When the gap lengths decay with enough regularity, the Cantor set Ca\mathcal{C}_a turns out to be the attractor of an iterated function system, and the smoothness of that system is governed precisely by the rate of decay: García [21] showed that the sets Cp\mathcal{C}_p associated with an=n−pa_n=n^{-p} are C1+1/pC^{1+1/p}-hyperbolic, with 1/p1/p the optimal Hölder exponent, and characterised which central Cantor sets admit realising systems of a given smoothness. This is not a separate story. The decay condition that makes Ca\mathcal{C}_a dynamically realisable is a strengthening of the condition 0<Hs(Ca)≤P0s(Ca)<∞0<\mathcal{H}^s(\mathcal{C}_a)\leq\mathcal{P}_0^s(\mathcal{C}_a)<\infty studied in § 3, so the dynamically defined sets are among the best-behaved members of the theory developed there. In the other direction, the dynamical description gives information the sequence alone does not: through Sullivan’s scale function, a complete smooth invariant, Cp\mathcal{C}_p is C1+1/pC^{1+1/p}-conjugate to a self-similar set, and this conjugacy transports the whole theory of Bernoulli convolutions to a genuinely non-linear family.

That transport is what makes the last topic of the survey accessible. Whether the arithmetic sum C+C′C + C' of two Cantor sets on the line contains an interval, or has positive measure, is a question with a long history in dynamics, going back to Newhouse’s work on homoclinic bifurcations and to the Palis conjecture. For the family {Cp}\{C_p\} both the thickness criterion and the transversality method of Peres and Solomyak can be adapted, giving dim⁡H(Cp+Cq)=min⁡{1/p+1/q,1}\dim_H(C_p + C_q) = \min\{1/p + 1/q, 1\} for almost every pp with qq fixed: an exact non-linear counterpart of the Peres–Shmerkin formula for self-similar sets.

The survey is organised as follows. § 2 collects the dimensions and measures used throughout: the Hausdorff, box, intermediate and Assouad-type dimensions, the Φ\Phi-dimensions, and dimension functions with the associated Hausdorff and packing hh-measures. § 3 introduces the family Ca\mathcal{C}_a of complementary sets, its Cantor subclass and the distinguished members CaC_a and DaD_a, and treats the question of what a rearrangement can and cannot change. § 4 restricts to the Cantor sets CaC_a and turns to their dynamical description, and § 5 to sums of Cantor sets and thickness. The two halves are largely independent: a reader interested only in rearrangements may stop at the end of § 3, and § 4 and § 5 use from it only the pp-Cantor sets of Example 3.2, Theorem 3.3 and the Hausdorff and packing hh-measures of § 3.3. Proofs are kept to a minimum, and open problems are stated where the picture is incomplete.

Preliminaries on dimensions

Throughout, ∣A∣|A| denotes the diameter of a non-empty set A⊂RnA \subset\mathbb{R}^n, and we write an≍bna_n \asymp b_n to mean that there exist constants c1,c2>0c_1, c_2 > 0, independent of nn, such that c1bn≤an≤c2bnc_1b_n \le a_n \le c_2b_n for all nn. Given δ>0\delta> 0, a family {Ui}i∈I\{U_i\}_{i\in I} is a δ\delta-cover of EE if E⊂⋃iUiE \subset\bigcup_i U_i and ∣Ui∣≤δ|U_i| \le\delta for all ii. All the sets we consider from § 3 onwards are subsets of [0,1][0,1], but the definitions below are stated in Rn\mathbb{R}^n for the sake of generality.

For E⊂RnE \subset\mathbb{R}^n, s>0s > 0 and δ>0\delta> 0, set

Hδs(E)=inf⁡{∑i=1∞∣Ui∣s:{Ui} is a δ-cover of E}.\mathcal{H}^s_\delta(E) = \inf\left\{ \sum_{i=1}^{\infty} |U_i|^s : \{U_i\}\text{ is a }\delta\text{-cover of }E \right\}.

The quantity Hδs(E)\mathcal{H}^s_\delta(E) is non-decreasing as δ\delta decreases, so the limit

Hs(E)=lim⁡δ→0Hδs(E)\mathcal{H}^s(E) = \lim_{\delta\to0} \mathcal{H}^s_\delta(E)

exists; it is the ss-dimensional Hausdorff measure of EE. If t>st > s and {Ui}\{U_i\} is a δ\delta-cover of EE, then ∑i∣Ui∣t≤δt−s∑i∣Ui∣s\sum_i |U_i|^t \le\delta^{t-s}\sum_i |U_i|^s, whence there is a critical exponent at which Hs(E)\mathcal{H}^s(E) jumps from ∞\infty to $0. The Hausdorff dimension of EE is

dim⁡HE=inf⁡{s≥0:Hs(E)=0}=sup⁡{s≥0:Hs(E)=∞}.\dim_H E = \inf\{s \ge0 : \mathcal{H}^s(E) = 0\} = \sup\{s \ge0 : \mathcal{H}^s(E) = \infty\}.

Equivalently, and in the form that will be relevant below,

dim⁡HE=inf⁡{s≥0:∀ ε>0 ∃ a cover {Ui} of E with ∑i∣Ui∣s<ε}.(1)\dim_H E = \inf\left\{s \ge0 : \forall\ \varepsilon> 0\ \exists\text{ a cover }\{U_i\}\text{ of }E\text{ with }\sum_i |U_i|^s < \varepsilon\right\}. \tag*{(1)}

We refer to [18] for the standard properties of Hs\mathcal{H}^{s} and dim⁡H\dim_{H}; we shall use repeatedly that countable sets have Hausdorff dimension 0.

The Hausdorff measure Hs\mathcal{H}^{s} assigns a finite positive value to EE for at most one exponent ss, and for many sets for none at all. Replacing the power xsx^{s} by a general gauge hh produces a finer scale of measures, and pairing it with the dual notion obtained from packings rather than covers gives the framework in which the sets of § 3 are best described: the pair refines the classification of Cantor sets by dimension functions (§ 3.3) and reveals an extremal reversal within Ca\mathcal{C}_{a} (§ 3.4) that is invisible at the level of Hausdorff measure alone.

A dimension function is a map h:(0,∞)→R+h:(0,\infty)\to\mathbb{R}^{+} that is increasing, continuous, doubling (that is, h(2x)≤ch(x)h(2x)\leq ch(x) for some constant cc) and satisfies lim⁡x→0h(x)=0\lim_{x\to0}h(x)=0. We write D\mathcal{D} for the class of all dimension functions, and gs(x)=xsg_{s}(x)=x^{s} for the power functions.

Definition 2.1. Given g,h∈Dg,h\in\mathcal{D} we write

g≺h ⟺ lim⁡t→0h(t)g(t)=0,g∼h ⟺ c1h(t)≤g(t)≤c2h(t)for small t,g\prec h \ \Longleftrightarrow\ \lim_{t\to0}\frac{h(t)}{g(t)}=0,\qquad g\sim h \ \Longleftrightarrow\ c_{1}h(t)\leq g(t)\leq c_{2}h(t)\quad\text{for small }t,

and g⪯hg\preceq h if g≺hg\prec h or g∼hg\sim h.

The direction of ≺\prec is worth emphasising: g≺hg\prec h means that hh is of smaller order than gg near the origin, so that gs≺gtg_{s}\prec g_{t} exactly when s<ts<t, and the ordering of gauges matches the ordering of the corresponding dimensions. The same relation controls Hausdorff and packing (pre)measures simultaneously, which is what allows estimates to be transferred from one to the other throughout.

Given E⊂RE\subset\mathbb{R} and δ>0\delta>0, a δ\delta-covering of EE is a countable family of sets of diameter at most δ\delta whose union contains EE, and a δ\delta-packing of EE is a disjoint family of open intervals of diameter at most δ\delta centred at points of EE. For h∈Dh\in\mathcal{D} set

Hδh(E)=inf⁡{∑i≥1h(∣Ei∣):{Ei} is a δ-covering of E},Hh(E)=lim⁡δ→0Hδh(E),\mathcal{H}^{h}_{\delta}(E)=\inf\left\{\sum_{i\geq1}h(|E_{i}|):\{E_{i}\}\text{ is a }\delta\text{-covering of }E\right\},\qquad \mathcal{H}^{h}(E)=\lim_{\delta\to0}\mathcal{H}^{h}_{\delta}(E),

and

Pδh(E)=sup⁡{∑i≥1h(∣Ei∣):{Ei} is a δ-packing of E},P0h(E)=lim⁡δ→0Pδh(E).\mathcal{P}^{h}_{\delta}(E)=\sup\left\{\sum_{i\geq1}h(|E_{i}|):\{E_{i}\}\text{ is a }\delta\text{-packing of }E\right\},\qquad \mathcal{P}^{h}_{0}(E)=\lim_{\delta\to0}\mathcal{P}^{h}_{\delta}(E).

The set function Hh\mathcal{H}^{h} is the Hausdorff hh-measure. The quantity P0h\mathcal{P}^{h}_{0}, the hh-packing premeasure, is not a measure, and one obtains the hh-packing measure by the usual regularisation

Ph(E)=inf⁡{∑i≥1P0h(Ei):E=⋃i≥1Ei}.\mathcal{P}^{h}(E)=\inf\left\{\sum_{i\geq1}\mathcal{P}^{h}_{0}(E_{i}):E=\bigcup_{i\geq1}E_{i}\right\}.

Always P0h(E)≤Ph(E)\mathcal{P}^{h}_{0}(E)\leq\mathcal{P}^{h}(E), and since hh is doubling, Hh(E)≤Ph(E)\mathcal{H}^{h}(E)\leq\mathcal{P}^{h}(E). Taking h=gsh=g_{s} recovers the usual ss-dimensional Hausdorff and packing (pre)measures, and the index inf⁡{s:P0gs(E)=0}\inf\{s:\mathcal{P}^{g_{s}}_{0}(E)=0\}—the pre-packing dimension of EE—coincides with dim⁡BE\dim_{B}E. The gap between P0h\mathcal{P}^{h}_{0} and Ph\mathcal{P}^{h} is not a technicality here: a countable set has zero packing measure for every hh, but may well have infinite premeasure.

Finally, EE is called an hh-set if 0<Hh(E)<∞0 < \mathcal{H}^{h}(E) < \infty, and hh is subadditive if h(x+y)≤C(h(x)+h(y))h(x+y) \leq C(h(x) + h(y)) for all x,yx,y and some constant CC.

Not every set is an hh-set; the simplest example is countable: if E={1/n:n∈N}∪{0}E = \{1/n : n \in\mathbb{N}\} \cup\{0\}, then Hh\mathcal{H}^{h} is non-atomic for every h∈Dh \in\mathcal{D}, since h(x)→0h(x) \to0 as x→0x \to0, so countable additivity forces Hh(E)=0\mathcal{H}^{h}(E) = 0 for every gauge. This example is unsatisfying, however, because EE is countable and hence of Hausdorff dimension 0 for trivial reasons. A genuinely fractal example is due to Best [11].

The same dichotomy appears for sets arising in Diophantine approximation. Olsen and Renfro [33] located the exact cut point between the gauges for which the set L\mathrm{L} of Liouville numbers has infinite (indeed non-σ\sigma-finite) Hausdorff hh-measure and those for which it has measure zero, so L\mathrm{L} is not dimensional; more recently, Moreira, Erazo and Angelini [30] obtained the analogous description for the irrationals whose Lagrange value equals 3, and showed that L\mathrm{L} is strictly the larger of the two on this finer scale: there are gauges hh with Hh(L)=∞\mathcal{H}^{h}(\mathrm{L}) = \infty and Hh(k−1(3))=0\mathcal{H}^{h}(k^{-1}(3)) = 0.

Cabrelli, Darji and Molter [13] placed this question in a broader setting by allowing measures other than Hausdorff measures. A compact set MM in a Polish group is H\mathcal{H}-visible if it is an hh-set for some h∈Dh \in\mathcal{D}, and visible if 0<μ(M)<∞0 < \mu(M) < \infty for some translation-invariant Borel measure μ\mu, not necessarily of Hausdorff type; MM is strongly invisible if every translation-invariant Borel measure either vanishes on MM or fails to be σ\sigma-finite there. Not being an hh-set for any hh is thus weaker than being strongly invisible, and Best’s example is not known to be strongly invisible. Davies had already exhibited a strongly invisible compact subset of R\mathbb{R} (see [15]); Cabrelli, Darji and Molter extend the construction to every uncountable abelian Polish group.

For a bounded set E⊂RnE \subset\mathbb{R}^{n}, let Nδ(E)N_{\delta}(E) denote the smallest number of closed balls of radius δ\delta needed to cover EE. The lower and upper box dimensions of EE are

dim⁡‾BE=lim inf⁡δ→0log⁡Nδ(E)−log⁡δ,dim⁡‾BE=lim sup⁡δ→0log⁡Nδ(E)−log⁡δ.\underline{\dim}_{B} E = \liminf_{\delta\to0} \frac{\log N_{\delta}(E)}{-\log\delta}, \qquad\overline{\dim}_{B} E = \limsup_{\delta\to0} \frac{\log N_{\delta}(E)}{-\log\delta}.

and when both coincide we write dim⁡BE\dim_{B} E for the common value. Comparing covers by balls, by grid cubes of side δ\delta and by arbitrary sets of diameter at most δ\delta changes Nδ(E)N_{\delta}(E) only by a multiplicative constant and a bounded rescaling of δ\delta, so Nδ(E)N_{\delta}(E) may equally be taken to be the number of grid cubes of side δ\delta meeting EE, the least number of sets of diameter at most δ\delta covering EE, or the largest number of disjoint balls of radius δ\delta centred in EE.

For bounded EE, the box dimensions admit a reformulation parallel to (1), in which the exponent is again characterised through covering sums rather than counting functions:

dim⁡‾BE=inf⁡{s≥0: ∀ε>0 ∀δ0>0 ∃δ∈(0,δ0) and a cover {Ui} of E with ∣Ui∣=δ ∀i such that ∑i∣Ui∣s<ε}(2)\underline{\dim}_{B} E = \inf\left\{ s \geq0 : \ \forall\varepsilon> 0 \ \forall\delta_{0} > 0 \ \exists\delta\in(0,\delta_{0}) \ \text{and a cover } \{U_i\} \text{ of } E \text{ with } |U_i| = \delta\ \forall i \text{ such that } \sum_i |U_i|^s < \varepsilon\right\} \tag*{(2)}

and

dim⁡‾BE=inf⁡{s≥0: ∀ε>0 ∃δ0>0 such that ∀δ∈(0,δ0) there is a cover {Ui} of E with ∣Ui∣=δ ∀i such that ∑i∣Ui∣s<ε}.(3)\overline{\dim}_B E=\inf\left\{s\ge0:\ \forall\varepsilon>0\ \exists\delta_0>0\ \text{such that}\ \forall\delta\in(0,\delta_0)\ \text{there is a cover }\{U_i\}\text{ of }E\text{ with }|U_i|=\delta\ \forall i\text{ such that }\sum_i |U_i|^s<\varepsilon\right\}. \tag*{(3)}

Two further dimensions will be needed, both of which describe the extreme local behaviour of a set rather than its global size. Given a non-empty E⊂RnE\subset\mathbb{R}^n, recall that Nr(E)N_r(E) denotes the smallest number of closed balls of radius rr needed to cover EE. The Assouad dimension of EE, is defined as follows

dim⁡AE=inf⁡{α:∃c,ρ>0 such that sup⁡x∈ENr(B(x,R)∩E)≤c(Rr)α for all 0<r<R<ρ}.\dim_A E=\inf\left\{\alpha:\exists c,\rho>0\text{ such that }\sup_{x\in E}N_r(B(x,R)\cap E)\le c\left(\frac{R}{r}\right)^\alpha\text{ for all }0<r<R<\rho\right\}.

and its dual, the lower Assouad dimension is

dim⁡LE=sup⁡{α:∃c,ρ>0 such that inf⁡x∈ENr(B(x,R)∩E)≥c(Rr)α for all 0<r<R<ρ}.\dim_L E=\sup\left\{\alpha:\exists c,\rho>0\text{ such that }\inf_{x\in E}N_r(B(x,R)\cap E)\ge c\left(\frac{R}{r}\right)^\alpha\text{ for all }0<r<R<\rho\right\}.

Whereas the Hausdorff dimension records global geometric information, these two quantities record the coarser but local behaviour of EE, since only one ball B(x,R)B(x,R) is examined at a time. They are correspondingly sensitive to local irregularities: if EE has an isolated point, then dim⁡LE=0\dim_L E=0. For a compact set EE they bracket all the dimensions considered above,

dim⁡LE≤dim⁡HE≤dim⁡‾BE≤dim⁡‾BE≤dim⁡AE.(4)\dim_L E\le\dim_H E\le\underline{\dim}_B E\le\overline{\dim}_B E\le\dim_A E. \tag*{(4)}

For further information on Assouad-like dimensions we refer the reader to Fraser’s book [20].

Hausdorff and box dimension formulations (1) and (2), differ in a single structural respect: the Hausdorff dimension places no constraint on the diameters of the covering sets, so that countable covers by sets of widely differing sizes are admissible, whereas the box dimensions force all the sets of the cover to have the same diameter δ\delta, and hence the cover to be finite. Since the class of admissible covers is larger in (1), the infimum is smaller, and one obtains at once

dim⁡HE≤dim⁡‾BE≤dim⁡‾BE,\dim_H E\le\underline{\dim}_B E\le\overline{\dim}_B E,

inequalities which may be strict.

The two definitions therefore appear as the extreme cases of a single scheme. This is the point of view adopted by Falconer, Fraser and Kempton [19], who interpolated between them by allowing the diameters of the covering sets to range over a geometric window governed by a parameter θ\theta.

Definition 2.2. Let E⊂RnE \subset\mathbb{R}^{n} be bounded and let 0≤θ≤10 \leq\theta\leq1. The lower θ\theta-intermediate dimension of EE is

dim⁡‾θE=inf⁡{s≥0:∀ε>0 ∀δ0>0 ∃0<δ≤δ0 and a cover {Ui} of E\underline{\dim}_{\theta} E=\inf\left\{s \geq0 : \forall\varepsilon>0\ \forall\delta_{0}>0\ \exists0<\delta\leq\delta_{0}\text{ and a cover }\{U_i\}\text{ of }E \right.
with δ1/θ≤∣Ui∣≤δ and ∑i∣Ui∣s≤ε},\left. \text{with }\delta^{1/\theta} \leq\lvert U_i\rvert\leq\delta\text{ and }\sum_i \lvert U_i\rvert^s \leq\varepsilon\right\},

and the upper θ\theta-intermediate dimension of EE is

dim⁡‾θE=inf⁡{s≥0:∀ε>0 ∃δ0>0 such that ∀0<δ≤δ0\overline{\dim}_{\theta} E=\inf\left\{s \geq0 : \forall\varepsilon>0\ \exists\delta_{0}>0\text{ such that }\forall0<\delta\leq\delta_{0} \right.
there is a cover {Ui} of E\text{there is a cover }\{U_i\}\text{ of }E
with δ1/θ≤∣Ui∣≤δ and ∑i∣Ui∣s≤ε}.\left. \text{with }\delta^{1/\theta} \leq\lvert U_i\rvert\leq\delta\text{ and }\sum_i \lvert U_i\rvert^s \leq\varepsilon\right\}.

When the two values agree we write dim⁡θE\dim_{\theta} E for the common value.

With the convention δ1/0=0\delta^{1/0}=0, Definition 2.2 recovers the two extremes:

dim⁡‾0E=dim⁡‾0E=dim⁡HE,dim⁡‾1E=dim⁡BE,dim⁡‾1E=dim⁡BE.\underline{\dim}_{0} E=\overline{\dim}_{0} E=\dim_{\mathrm{H}} E,\qquad \underline{\dim}_{1} E=\dim_{\mathrm{B}} E,\qquad \overline{\dim}_{1} E=\dim_{\mathrm{B}} E.

For θ∈[0,1]\theta\in[0,1] the admissible covers are more numerous than those allowed in (2) and fewer than those allowed in (1), so that

dim⁡HE≤dim⁡‾θE≤dim⁡‾θE≤dim⁡BEanddim⁡‾θE≤dim⁡‾θE≤dim⁡BE,\dim_{\mathrm{H}} E \leq\underline{\dim}_{\theta} E \leq\overline{\dim}_{\theta} E \leq\dim_{\mathrm{B}} E \qquad\text{and}\qquad \underline{\dim}_{\theta} E \leq\overline{\dim}_{\theta} E \leq\dim_{\mathrm{B}} E,

and, by the same reasoning, θ↦dim⁡‾θE\theta\mapsto\underline{\dim}_{\theta} E and θ↦dim⁡‾θE\theta\mapsto\overline{\dim}_{\theta} E are non-decreasing, though not necessarily strictly increasing.

We record the properties that will be used later. Both dimensions are monotone under inclusion and invariant under bi-Lipschitz maps, and satisfy dim⁡‾θE=dim⁡‾θE\underline{\dim}_{\theta} E=\underline{\dim}_{\theta} E and dim⁡‾θE=dim⁡‾θE\overline{\dim}_{\theta} E=\overline{\dim}_{\theta} E for θ∈(0,1]\theta\in(0,1]. The upper intermediate dimension is finitely stable,

dim⁡‾θ(E∪E′)=max⁡{dim⁡‾θE,dim⁡‾θE′},\overline{\dim}_{\theta}(E \cup E')=\max\{\overline{\dim}_{\theta} E,\overline{\dim}_{\theta} E'\},

whereas dim⁡‾θ\underline{\dim}_{\theta} is not finitely stable, and neither dimension is countably stable. It was shown in [19] that θ↦dim⁡‾θE\theta\mapsto\underline{\dim}_{\theta} E and θ↦dim⁡‾θE\theta\mapsto\overline{\dim}_{\theta} E are continuous on (0,1](0,1]; continuity at θ=0\theta=0, however, may fail, so that the intermediate dimensions need not interpolate all the way down to dim⁡HE\dim_{\mathrm{H}} E.

These dimensions have proved to be more than an interpolation device. Being a function rather than a number, θ↦dim⁡θE\theta\mapsto\dim_{\theta} E is a finer bi-Lipschitz invariant than its endpoints: Banaji and Kolossváry [6] computed it exactly for Bedford–McMullen carpets and used it to obtain necessary conditions for Lipschitz equivalence. Its behaviour as θ\theta tends to 00 also characterises when the box dimension of a typical orthogonal projection attains its maximal value [1]. Knowing the whole spectrum therefore says more than knowing its two endpoints.

Example 2.3. Let p>0p>0 and Fp={0}∪{n−p:n∈N}F_p=\{0\}\cup\{n^{-p}:n\in\mathbb{N}\}. Since FpF_p is countable, dim⁡HFp=0\dim_{\mathrm{H}}F_p=0, while dim⁡BFp=1/(p+1)\dim_{\mathrm{B}}F_p=1/(p+1). It was proved in [19] that

dim⁡‾θFp=dim⁡‾θFp=θp+θ,θ∈(0,1].\underline{\dim}_{\theta}F_p=\overline{\dim}_{\theta}F_p=\frac{\theta}{p+\theta},\qquad\theta\in(0,1].

Here the interpolation is complete: the dimension function is continuous on all of [0,1][0,1] and tends to dim⁡HFp=0\dim_H F_p = 0 as θ→0+\theta\to0^+. In the language of § 3, FpF_p is the decreasing rearrangement DaD_a of the sequence an=n−p−(n+1)−pa_n = n^{-p} - (n+1)^{-p}, and the formula above is the case dim⁡‾BCa=1/(p+1)\overline{\dim}_B C_a = 1/(p+1) of Theorem 3.15 (ii).

Beyond these general properties, one may ask which functions of θ\theta occur as intermediate dimensions. The answer, obtained by Banaji and Rutar [7], is a complete characterisation, and the constraint turns out to be governed by the two Assouad dimensions of the set.

The starting point is a quantitative bound: the intermediate dimensions cannot grow arbitrarily fast in θ\theta.

Proposition 2.4 ([7 Cor. 2.8]). If dim⁡LE<dim⁡AE\dim_L E < \dim_A E, then for every θ∈(0,1]\theta\in(0,1],

dim⁡‾θE≥θdim⁡AE(dim⁡‾BE−dim⁡LE)+dim⁡LE(dim⁡AE−dim⁡‾BE)θ(dim⁡‾BE−dim⁡LE)+(dim⁡AE−dim⁡‾BE).\underline{\dim}_{\theta} E \ge\frac{\theta\dim_A E \left(\overline{\dim}_B E - \dim_L E\right) + \dim_L E \left(\dim_A E - \overline{\dim}_B E\right)}{\theta\left(\overline{\dim}_B E - \dim_L E\right) + \left(\dim_A E - \overline{\dim}_B E\right)}.

and the same holds with dim⁡‾θ\underline{\dim}_{\theta} and dim⁡‾B\overline{\dim}_B replaced by dim⁡‾θ\overline{\dim}_{\theta} and dim⁡B\dim_B.

Writing λ=dim⁡LE\lambda= \dim_L E, α=dim⁡AE\alpha= \dim_A E and β=dim⁡‾BE\beta= \overline{\dim}_B E, the right-hand side is exactly the solution of

h′(θ)=(h(θ)−λ)(α−h(θ))(α−λ)θ,h(1)=β.h'(\theta) = \frac{(h(\theta)-\lambda)(\alpha-h(\theta))}{(\alpha-\lambda)\theta}, \qquad h(1)=\beta.

A dimension profile therefore cannot decrease faster, as θ\theta decreases, than the extremal solution of this equation. This makes it natural to single out the functions satisfying the corresponding differential inequality.

Definition 2.5. For 0≤λ≤α≤n0 \le\lambda\le\alpha\le n, let B(λ,α)\mathcal{B}(\lambda,\alpha) be the class of functions h:[0,1]→[λ,α]h : [0,1] \to[\lambda,\alpha] that are non-decreasing, continuous on (0,1](0,1], and satisfy

lim sup⁡ϵ→0+h(θ+ϵ)−h(θ)ϵ≤(h(θ)−λ)(α−h(θ))(α−λ)θ\limsup_{\epsilon\to0^+} \frac{h(\theta+\epsilon)-h(\theta)}{\epsilon} \le\frac{(h(\theta)-\lambda)(\alpha-h(\theta))}{(\alpha-\lambda)\theta}

for all θ∈(0,1)\theta\in(0,1).

If λ=α\lambda=\alpha we set B(λ,λ)={h≡λ}\mathcal{B}(\lambda,\lambda)=\{h\equiv\lambda\}.

Theorem 2.6 ([7]). Let E⊂RnE \subset\mathbb{R}^n be bounded with dim⁡LE=λ\dim_L E=\lambda and dim⁡AE=α\dim_A E=\alpha. Then the functions h1(θ)=dim⁡‾θEh_1(\theta)=\underline{\dim}_{\theta}E and h2(θ)=dim⁡‾θEh_2(\theta)=\overline{\dim}_{\theta}E satisfy

h1,h2∈B(λ,α),h1≤h2,h1(0)=h2(0).h_1,h_2 \in\mathcal{B}(\lambda,\alpha), \qquad h_1 \le h_2, \qquad h_1(0)=h_2(0).

Conversely, given h1,h2∈B(λ,α)h_1,h_2 \in\mathcal{B}(\lambda,\alpha) with h1≤h2h_1 \le h_2 and h1(0)=h2(0)h_1(0)=h_2(0), there is a compact perfect set E⊂RnE \subset\mathbb{R}^n with

dim⁡LE=λ,dim⁡‾θE=h1(θ),dim⁡‾θE=h2(θ),dim⁡AE=α.\dim_L E=\lambda, \qquad\underline{\dim}_{\theta}E=h_1(\theta), \qquad\overline{\dim}_{\theta}E=h_2(\theta), \qquad\dim_A E=\alpha.

Theorem 2.6 settles the question for arbitrary compact sets. The corresponding question inside a fixed family Ca\mathcal{C}_a is open, and we return to it at the end of § 3.

Finally, we mention two extensions of this circle of ideas. In [5], Banaji introduced the Φ\Phi-intermediate dimensions, in which the admissible diameters are constrained to [Φ(δ),δ][\Phi(\delta),\delta], recovering the θ\theta-intermediate dimensions for Φ(x)=x1/θ\Phi(x)=x^{1/\theta} and extending the theory to general metric spaces and in [3], intermediate dimensions were defined for finite Borel measures, interpolating between the upper Hausdorff and the Minkowski dimensions of a measure.

Just as the intermediate dimensions interpolate between dim⁡H\dim_H and dim⁡B\dim_B, Lü and Xi [29] introduced a spectrum interpolating between the upper box dimension and the Assouad dimension. For non-empty E⊂RnE \subset\mathbb{R}^n and θ∈(0,1]\theta\in(0,1],

dim⁡AθE=inf⁡{α:∃C>0 such thatsup⁡x∈ENr(B(x,R)∩E)≤C(Rr)α for all 0<r≤R1/θ<R<1}.\dim_A^\theta E = \inf\left\{ \alpha: \exists C > 0 \text{ such that} \sup_{x \in E} N_r(B(x,R) \cap E) \le C\left(\frac{R}{r}\right)^\alpha \text{ for all } 0 < r \le R^{1/\theta} < R < 1 \right\}.

so that dim⁡A1E=dim⁡AE\dim_A^1 E = \dim_A E. The quasi-Assouad dimension is then defined as dim⁡qAE=lim⁡θ→1dim⁡AθE\dim_{qA} E = \lim_{\theta\to1} \dim_A^\theta E. In general

dim⁡‾BE≤dim⁡AθE≤dim⁡qAE≤dim⁡AE,\overline{\dim}_B E \le\dim_A^\theta E \le\dim_{qA} E \le\dim_A E,

and all the inequalities may be strict.

The Assouad dimensions admit refinements that interpolate towards the box dimensions, and these are of a different nature from the intermediate dimensions. Whereas dim⁡θ\dim_\theta and dim⁡‾θ\overline{\dim}_\theta restrict the global range of diameters admitted in a cover, the refinements below localise around each point and each scale, in the spirit of dim⁡A\dim_A itself. The most general such family is the one introduced by García, Hare and Mendivil [24, 25].

Definition 2.7. An admissible function is a map Φ:(0,1)→R+\Phi: (0,1) \to\mathbb{R}^+ such that R1+Φ(R)R^{1+\Phi(R)} decreases as R↓0R \downarrow0; in particular R1+Φ(R)≤RR^{1+\Phi(R)} \le R. Given a bounded, non-empty E⊂RnE \subset\mathbb{R}^n, the upper Φ\Phi-dimension of EE is

dim⁡‾ΦE=inf⁡{α:∃c1,c2>0 such thatsup⁡x∈ENr(B(x,R)∩E)≤c2(Rr)α for all 0<r≤R1+Φ(R)≤R<c1}.\overline{\dim}_{\Phi} E = \inf\left\{ \alpha: \exists c_1,c_2 > 0 \text{ such that} \sup_{x \in E} N_r(B(x,R) \cap E) \le c_2\left(\frac{R}{r}\right)^\alpha \text{ for all } 0 < r \le R^{1+\Phi(R)} \le R < c_1 \right\}.

and the lower Φ\Phi-dimension is

dim⁡‾ΦE=sup⁡{α:∃c1,c2>0 such thatinf⁡x∈ENr(B(x,R)∩E)≥c2(Rr)α for all 0<r≤R1+Φ(R)≤R<c1}.\underline{\dim}_{\Phi} E = \sup\left\{ \alpha: \exists c_1,c_2 > 0 \text{ such that} \inf_{x \in E} N_r(B(x,R) \cap E) \ge c_2\left(\frac{R}{r}\right)^\alpha \text{ for all } 0 < r \le R^{1+\Phi(R)} \le R < c_1 \right\}.

The choice Φ≡0\Phi\equiv0 recovers dim⁡AE\dim_A E and dim⁡LE\dim_L E. The constant function Φ≡1/θ−1\Phi\equiv1/\theta- 1, θ∈(0,1)\theta\in(0,1), gives the Assouad spectrum of Lü and Xi, for which we write dim⁡AθE=dim⁡‾ΦE\dim_A^\theta E = \overline{\dim}_{\Phi} E; letting Φ(R)→0\Phi(R) \to0 as R→0R \to0 yields the quasi-Assouad dimensions,

dim⁡qAE=lim⁡θ→1dim⁡AθE,\dim_{qA} E = \lim_{\theta\to1} \dim_A^\theta E,

and their lower counterpart dim⁡qLE\dim_{qL} E. For compact EE one has

dim⁡LE≤dim⁡qLE≤dim⁡HE≤dim⁡‾BE≤dim⁡‾BE≤dim⁡qAE≤dim⁡AE,\dim_{L} E \le\dim_{qL} E \le\dim_{H} E \le\underline{\dim}_{B} E \le\overline{\dim}_{B} E \le\dim_{qA} E \le\dim_{A} E,
dim⁡LE≤dim⁡ΦE≤dim⁡‾BE≤dim⁡‾BE≤dim⁡‾ΦE≤dim⁡AE,\dim_{L} E \le\dim_{\Phi} E \le\underline{\dim}_{B} E \le\overline{\dim}_{B} E \le\overline{\dim}_{\Phi} E \le\dim_{A} E,

and all the inequalities may be strict.

Cut-out sets and their rearrangements

Let a={an}a = \{a_n\} be a positive, non-increasing, summable sequence with sum 11, and let Ia=[0,1]I_a = [0,1]. We denote by Ca\mathcal{C}_a the family of all compact sets E⊂IaE \subset I_a of the form E=Ia∖⋃j≥1UjE = I_a \setminus\bigcup_{j\ge1} U_j, where {Uj}\{U_j\} is a pairwise disjoint family of open intervals contained in IaI_a with ∣Uj∣=aj|U_j| = a_j for every jj. The sets in Ca\mathcal{C}_a are called the complementary sets of aa. Every E∈CaE \in\mathcal{C}_a is compact, totally disconnected and of Lebesgue measure zero, and contains the endpoints 00 and 11; conversely, every infinite compact subset of [0,1][0,1] of measure zero containing 00 and 11 belongs to Ca\mathcal{C}_a for a unique such sequence aa. Note that Ca\mathcal{C}_a depends only on the collection of lengths {aj}\{a_j\} and not on the order in which they are listed, so no generality is lost in assuming aa non-increasing.

Two members of Ca\mathcal{C}_a will play a distinguished role. The first is the countable set

Da={0}∪{∑i≥kai:k∈N},D_a = \{0\} \cup\left\{ \sum_{i\ge k} a_i : k \in\mathbb{N} \right\},

a decreasing sequence of points accumulating at 00, whose consecutive gaps are exactly the terms of aa arranged in decreasing order. We call DaD_a the decreasing rearrangement of aa.

The second is the Cantor set associated with aa, denoted CaC_a and constructed inductively as follows. In the first step, remove from IaI_a an open interval of length a1a_1, leaving two closed intervals I1,1I_{1,1} and I1,2I_{1,2}. Having completed the kk-th step, we have 2k2^k closed intervals Ik,1,…,Ik,2kI_{k,1},\ldots,I_{k,2^k}; from each Ik,jI_{k,j} we remove an open interval of length a2k+j−1a_{2^k+j-1}, obtaining the two closed intervals Ik+1,2j−1I_{k+1,2j-1} and Ik+1,2jI_{k+1,2j}. We then set

Ca:=⋂k≥1⋃j=12kIk,j.C_a := \bigcap_{k\ge1} \bigcup_{j=1}^{2^k} I_{k,j}.

The construction determines CaC_a uniquely, since the lengths of the gaps removed on either side of a given gap are prescribed; for instance, the interval of length a1a_1 must be positioned so that the remaining closed interval on its left has length a2+a4+a5+a8+⋯a_2+a_4+a_5+a_8+\cdots. The classical middle-third Cantor set is CaC_a for the sequence given by aj=3−na_j = 3^{-n} when 2n−1≤j<2n2^{n-1} \le j < 2^n.

Since every E∈CaE \in\mathcal{C}_a is already compact, totally disconnected and of measure zero, such a set is homeomorphic to the classical Cantor set precisely when it is perfect. We shall accordingly refer to a perfect E∈CaE \in\mathcal{C}_a as a Cantor rearrangement of aa, reserving the notation CaC_a for the particular set constructed above. Not every complementary set is of this type: DaD_a consists entirely of isolated points except for 00.

Given a non-increasing, summable sequence a={an}a=\{a_n\}, define

xn(a)=∑i=n∞aiandsn(a)=2−nx2n(a).x_n(a)=\sum_{i=n}^{\infty}a_i \quad\text{and}\quad s_n(a)=2^{-n}x_{2^n}(a).

When the sequence aa is fixed and clear from context, we will simply write sns_n instead of sn(a)s_n(a).

Note that since aa is non-increasing we have the estimate

sn+1<∣In,j∣<sn−1for all 1≤j≤2n,s_{n+1}<|I_{n,j}|<s_{n-1}\quad\text{for all }1\leq j\leq2^n,

where In,jI_{n,j} are the closed intervals of step nn in the construction of the Cantor set CaC_a.

Hausdorff measure and dimension

The Hausdorff dimension of the sets in Ca\mathcal C_a was determined by Besicovitch and Taylor [10], who showed that CaC_a is extremal in the family and that every smaller value is attained. Throughout we write α(a):=dim⁡HCa\alpha(a):=\dim_H C_a.

Theorem 3.1 ([10]). Let a={an}a=\{a_n\} be a positive, non-increasing, summable sequence. Then

α(a)=lim inf⁡n→∞log⁡n−log⁡(xn/n).\alpha(a)=\liminf_{n\to\infty}\frac{\log n}{-\log(x_n/n)}.

and dim⁡HE≤α(a)\dim_H E\leq\alpha(a) for every E∈CaE\in\mathcal C_a. Conversely, for every β∈[0,α(a)]\beta\in[0,\alpha(a)] there is a set E∈CaE\in\mathcal C_a with dim⁡HE=β\dim_H E=\beta. If moreover α(a)>0\alpha(a)>0, the measure itself can be prescribed: for every β∈[0,α(a)]\beta\in[0,\alpha(a)] and every 0≤γ≤Λ(β,a)0\leq\gamma\leq\Lambda(\beta,a) there is E∈CaE\in\mathcal C_a with dim⁡HE=β\dim_H E=\beta and Hβ(E)=γ\mathcal H^\beta(E)=\gamma, where

Λ(β,a):=lim inf⁡n→∞n(xnn)β.\Lambda(\beta,a):=\liminf_{n\to\infty}n\left(\frac{x_n}{n}\right)^\beta.

Thus {dim⁡HE:E∈Ca}=[0,α(a)]\{\dim_H E:E\in\mathcal C_a\}=[0,\alpha(a)], the minimum being attained by the decreasing rearrangement DaD_a, which is countable and hence has Hausdorff dimension 00, and the maximum by CaC_a. The quantity Λ(β,a)\Lambda(\beta,a) reappears in § 3.4 as the case h=gβh=g_\beta of the estimates of García, Molter and Scotto for Hh(Ca)\mathcal H^h(C_a).

It was shown in [26] that the two lower dimensions agree on CaC_a,

dim⁡HCa=dim⁡BCa,\dim_H C_a=\dim_B C_a,

which is what makes the interpolation degenerate from below in Theorem 3.15 (i).

Example 3.2 (pp-Cantor sets). Let p>1p>1 and let λ={λk}\lambda=\{\lambda_k\} with λk=k−p\lambda_k=k^{-p}, normalised so that ∑kλk=1\sum_k\lambda_k=1. Then xn≍n1−px_n\asymp n^{1-p} and Theorem 3.1 gives dim⁡HCλ=1/p\dim_H C_\lambda=1/p. Consequently dim⁡HE≤1/p\dim_H E\leq1/p for every E∈CλE\in\mathcal C_\lambda, and for each s∈[0,1/p]s\in[0,1/p] there is a rearrangement E∈CλE\in\mathcal C_\lambda with dim⁡HE=s\dim_H E=s and Hs(E)>0\mathcal H^s(E)>0.

Dimensionality and visibility

Passing from a single exponent ss to a general gauge hh enlarges considerably the class of sets that can be assigned a finite positive size, but it does not exhaust it. Call a compact set EE dimensional if EE is an hh-set for some h∈Dh\in\mathcal D. Not every Cantor set CaC_a is an ss-set. What is true, and this is the point of the following theorem, is that a suitable gauge always exists.

Theorem 3.3 ([14]). Let a={ak}a = \{a_k\} be positive, non-increasing, with ∑kak=1\sum_k a_k = 1, and let CaC_a be the associated Cantor set. Then there is a dimension function h∈Dh \in\mathcal{D} for which CaC_a is an hh-set; in fact hh may be chosen concave, and then

132≤Hh(Ca)≤1.\frac{1}{32} \leq\mathcal{H}^{h}(C_a) \leq1.

The gauge function is constructed explicitly from the sequence. Any h∈Dh \in\mathcal{D} with h(xn/n)∼1/nh(x_n/n) \sim1/n is called an associated dimension function of aa, and we write hah_a for any such gauge; all of them are comparable, so hah_a is well defined up to ∼\sim.

Packing measures and the classification of gauges

The two orderings on D\mathcal{D} translate directly into statements about measures, and the translation is the same for Hausdorff measure, packing premeasure and packing measure.

Lemma 3.4 ([26]). Let νg\nu^g denote any of Hg\mathcal{H}^g, P0g\mathcal{P}_0^g or Pg\mathcal{P}^g, for g∈Dg \in\mathcal{D}, and let E⊂RE \subset\mathbb{R}.

(i) If f≺hf \prec h and νf(E)<∞\nu^f(E) < \infty, then νh(E)=0\nu^h(E) = 0.

(ii) If f∼hf \sim h, then νf(E)<∞\nu^f(E) < \infty if and only if νh(E)<∞\nu^h(E) < \infty, and νf(E)>0\nu^f(E) > 0 if and only if νh(E)>0\nu^h(E) > 0. In particular EE is an ff-set precisely when it is an hh-set, and likewise for the packing analogues.

The Besicovitch–Taylor estimates for Hs(Ca)\mathcal{H}^s(C_a) extend to arbitrary gauges, and admit an exact packing dual.

Theorem 3.5 ([26 Prop. 4.1, Thm. 4.2]). Let aa be positive, non-increasing and summable, and put bn:=xn/nb_n := x_n/n. Then for every h∈Dh \in\mathcal{D},

14lim inf⁡n→∞nh(bn)≤Hh(Ca)≤4lim inf⁡n→∞nh(bn),\frac{1}{4}\liminf_{n\to\infty} n h(b_n) \leq\mathcal{H}^{h}(C_a) \leq4\liminf_{n\to\infty} n h(b_n),
18lim sup⁡n→∞nh(bn)≤P0h(Ca)≤8lim sup⁡n→∞nh(bn).\frac{1}{8}\limsup_{n\to\infty} n h(b_n) \leq\mathcal{P}_0^{h}(C_a) \leq8\limsup_{n\to\infty} n h(b_n).

The two estimates differ only in the replacement of lim inf⁡\liminf by lim sup⁡\limsup, and of covering by packing. Since an associated dimension function satisfies ha(bn)∼1/nh_a(b_n) \sim1/n, both quantities nha(bn)n h_a(b_n) are bounded above and below, so Theorem 3.5 gives at once that Hha(Ca)\mathcal{H}^{h_a}(C_a) and P0ha(Ca)\mathcal{P}_0^{h_a}(C_a) are finite and positive: the same gauge regularises the covering and the packing process simultaneously.

Combining the two halves characterises the gauges that measure CaC_a, and in particular tells when the associated dimension function reduces to a power.

Theorem 3.6 ([26]). Let aa be non-increasing and let h∈Dh \in\mathcal{D} satisfy 0<Hh(Ca)≤P0h(Ca)<∞0 < \mathcal{H}^{h}(C_a) \leq\mathcal{P}_0^{h}(C_a) < \infty. Then, for every g∈Dg \in\mathcal{D},

g∼h⟺0<Hg(Ca)≤P0g(Ca)<∞,g \sim h \Longleftrightarrow0 < \mathcal{H}^{g}(C_a) \leq\mathcal{P}_0^{g}(C_a) < \infty,
g≺h⟺Hg(Ca)=+∞,h≺g⟺P0g(Ca)=0.g \prec h \Longleftrightarrow\mathcal{H}^{g}(C_a) = +\infty,\qquad h \prec g \Longleftrightarrow\mathcal{P}_0^{g}(C_a) = 0.

In particular h∼gsh \sim g_s if and only if 0<Hs(Ca)≤P0s(Ca)<∞0 < \mathcal{H}^{s}(C_a) \leq\mathcal{P}_0^{s}(C_a) < \infty.

Being an ss-set is therefore not enough for the associated dimension function to be equivalent to a power: the condition 0<Hs(Ca)<∞0 < \mathcal{H}^{s}(C_a) < \infty must be supplemented by P0s(Ca)<∞\mathcal{P}_0^{s}(C_a) < \infty. That the extra condition is not automatic is shown by the following.

Example 3.7 ([26]). For every s∈(0,1)s \in(0,1) there is a non-increasing sequence aa, obtained by perturbing a geometric sequence as λk=2−k/(s+εk)\lambda_k = 2^{-k/(s+\varepsilon_k)} with {εk}\{\varepsilon_k\} a null sequence supported on alternate dyadic blocks, such that

dim⁡HCa=dim⁡BCa=dim⁡‾BCa=s,0<Hs(Ca)<∞,P0s(Ca)=+∞.\dim_H C_a = \dim_B C_a = \overline{\dim}_B C_a = s,\qquad0 < \mathcal{H}^s(C_a) < \infty,\qquad\mathcal{P}_0^s(C_a) = +\infty.

Thus CaC_a is an ss-set whose box dimension exists, and yet ha≁gsh_a \not\sim g_s.

The gauges therefore classify these sets more finely than the exponents alone. For p>1p > 1 and q∈Rq \in\mathbb{R}, the sequence an=(log⁡n)qn−pa_n = (\log n)^q n^{-p}, n>1n > 1, produces Cantor sets Cp,qC_{p,q} with dim⁡HCp,q=dim⁡BCp,q=1/p\dim_H C_{p,q} = \dim_B C_{p,q} = 1/p irrespective of qq, while

hp,q(x):=x1/p(−log⁡x)−q/ph_{p,q}(x) := x^{1/p}(-\log x)^{-q/p}

satisfies 0<Hhp,q(Cp,q)≤P0hp,q(Cp,q)<∞0 < \mathcal{H}^{h_{p,q}}(C_{p,q}) \leq\mathcal{P}_0^{h_{p,q}}(C_{p,q}) < \infty, see [26 Corollary 4.8]. Ordering the pairs (1/p,q)(1/p,q) lexicographically, the family {hp,q}\{h_{p,q}\} separates sets that the power functions {gs}\{g_s\} cannot tell apart.

Hausdorff and packing hh-measures of rearrangements

The results of Section 3.3 concern the single set CaC_a. We now turn to the whole family, where the natural question is how Hh\mathcal{H}^h and Ph\mathcal{P}^h vary across Ca\mathcal{C}_a for a fixed gauge. Hare, Mendivil and Zuberman [27] extended both the Besicovitch–Taylor extremality statement and its packing dual to arbitrary h∈Dh \in\mathcal{D}, and found that the two behave in opposite ways.

The statements below are sharper than mere attainability of values: they locate the realising sets inside Ca\mathcal{C}_a.

Proposition 3.8 ([27]). For h∈Dh \in\mathcal{D}, there is a constant cc, depending only on the doubling constant of hh, such that Hh(E)≤cHh(Ca)\mathcal{H}^h(E) \leq c\mathcal{H}^h(C_a) for every E∈CaE \in\mathcal{C}_a. That is, CaC_a has, up to a constant, the maximal Hausdorff hh-measure within Ca\mathcal{C}_a, generalising the Besicovitch–Taylor extremality result beyond power functions gsg_s. When CaC_a is a middle-τ\tau Cantor set the constant may be taken to be c=1c = 1, so CaC_a is then exactly maximal; in general the sharp constant and the identity of a maximiser (if any exists) remain open.

Theorem 3.9 ([27]). Let h≺hah \prec h_a, where hah_a is the associated dimension function of CaC_a, and let γ≥0\gamma\geq0. There is a perfect set E∈CaE \in\mathcal{C}_a with Hh(E)=γ\mathcal{H}^h(E) = \gamma. Moreover the set {E∈Ca:Hh(E)=γ}\{E \in\mathcal{C}_a : \mathcal{H}^h(E) = \gamma\} is dense in Ca\mathcal{C}_a with respect to the Hausdorff metric ρ\rho.

Theorem 3.9 sharpens the Besicovitch–Taylor attainability statement—that any β∈[0,α(a)]\beta\in[0,\alpha(a)] is realised as dim⁡HE\dim_H E for some E∈CaE \in\mathcal{C}_a—to the level of hh-measures rather than merely dimension values, and it guarantees that the realising set can always be chosen perfect, and in fact densely so, within (Ca,ρ)(\mathcal{C}_a,\rho).

The dual statement for packing measure exhibits the opposite extremal pattern, dual to Proposition 3.8.

Proposition 3.10 ([27]). There is a constant cc such that P0h(Ca)≤cP0h(E)\mathcal{P}_0^h(C_a) \leq c\mathcal{P}_0^h(E) for every E∈CaE \in\mathcal{C}_a and every h∈Dh \in\mathcal{D}. That is, despite having the maximal Hausdorff hh-measure by Proposition 3.8, CaC_a has—again up to a constant—the minimal packing hh-premeasure within Ca\mathcal{C}_a. As with the Hausdorff case, the sharp constant and

the exact minimising set are unknown in general; for the middle-third Cantor set, Ps(Ca)=4s≠lim sup⁡nn(xn/n)s\mathcal{P}^{s}(C_a)=4^s \ne\limsup_{n} n(x_n/n)^s where s=1s=1 where s=log⁡2/log⁡3s=\log2/\log3, so the elementary bound is not sharp even in this classical case.

Theorem 3.11 ([27]). Let h∈Dh \in\mathcal{D} be subadditive (e.g. any hh equivalent to a concave function, which includes every associated dimension function hah_a). The following are equivalent:

(i) some E∈CaE \in\mathcal{C}_a has P0h(E)>0\mathcal{P}_0^h(E)>0;

(ii) P0h(Da)=∞\mathcal{P}_0^h(D_a)=\infty;

(iii) ∑nh(an)=∞\sum_n h(a_n)=\infty;

(iv) some perfect E∈CaE \in\mathcal{C}_a has Ph(E)=∞\mathcal{P}^h(E)=\infty.

Consequently {Ph(E):E∈Ca}\{\mathcal{P}^h(E):E\in\mathcal{C}_a\} is either {0}\{0\} or [0,∞][0,\infty]: there is no intermediate regime for the family of packing hh-measures attained across Ca\mathcal{C}_a.

In particular, whenever P0ha(Ca)>0\mathcal{P}^{h_a}_0(C_a)>0—which always holds, since hah_a is an associated dimension function of CaC_a—there exists a rearrangement E∈CaE\in\mathcal{C}_a with infinite hah_a-packing measure, even though DaD_a itself trivially has zero packing measure (being countable) and CaC_a realises the minimal packing premeasure in the family.

Theorem 3.12 ([27]). Let h≼hah\preccurlyeq h_a and γ>0\gamma>0. There is a perfect set E∈CaE\in\mathcal{C}_a with Ph(E)=γ\mathcal{P}^h(E)=\gamma, and the set {E∈Ca:Ph(E)=γ}\{E\in\mathcal{C}_a:\mathcal{P}^h(E)=\gamma\} is dense in (Ca,ρ)(\mathcal{C}_a,\rho).

Remark 3.13. Propositions 3.8 and 3.10, taken together, show that the roles of CaC_a and the general complementary sets E∈CaE\in\mathcal{C}_a are exactly reversed between the Hausdorff and packing measures: CaC_a maximises the former and minimises the latter (up to uniform constants), while it is the more disordered rearrangements E∈CaE\in\mathcal{C}_a that can achieve arbitrarily small Hausdorff hh-measure or arbitrarily large—even infinite—packing hh-measure.

Box dimension

The box dimensions of sets in Ca\mathcal{C}_a were carefully studied in [17 Chapter 3.2]. There, Falconer shows that the upper and lower box dimensions of sets in Ca\mathcal{C}_a depend only on the sequence aa, and not on the rearrangement. That is, given any complementary set, its lower box dimension equals that of every other complementary set, and the same holds for the upper box dimension. Furthermore, for any E∈CaE\in\mathcal{C}_a,

lim inf⁡n→∞log⁡n−log⁡an≤dim⁡‾BE≤dim⁡‾BE≤lim sup⁡n→∞log⁡n−log⁡an,\liminf_{n\to\infty}\frac{\log n}{-\log a_n}\leq\underline{\dim}_B E\leq\overline{\dim}_B E\leq\limsup_{n\to\infty}\frac{\log n}{-\log a_n},

and the box dimension exists if and only if these limits coincide.

The upper and lower box dimension of mathcalCamathcal{C}_a (and therefore of any other complementary set in Ca\mathcal{C}_a) are given by

dim⁡‾BCa=lim sup⁡n→∞log⁡n−log⁡(xn/n),dim⁡‾BCa=lim inf⁡n→∞log⁡n−log⁡(xn/n).\overline{\dim}_B C_a=\limsup_{n\to\infty}\frac{\log n}{-\log(x_n/n)},\qquad\underline{\dim}_B C_a=\liminf_{n\to\infty}\frac{\log n}{-\log(x_n/n)}.

Moreover, it was shown in [42] that

lim sup⁡n→∞log⁡n−log⁡(xn/n)=lim sup⁡n→∞log⁡n−log⁡an.\limsup_{n\to\infty}\frac{\log n}{-\log(x_n/n)}=\limsup_{n\to\infty}\frac{\log n}{-\log a_n}.

This equivalence does not hold for the lower box dimension. In fact, the inequality

lim inf⁡n→∞log⁡n−log⁡an≤lim inf⁡n→∞log⁡n−log⁡(xn/n)\liminf_{n\to\infty}\frac{\log n}{-\log a_n}\leq\liminf_{n\to\infty}\frac{\log n}{-\log(x_n/n)}

can be strict; see, for instance, [14 Proposition 5].

Intermediate dimension

In order to compute the θ\theta-intermediate dimension of complementary sets, Angelini and Molter [2] introduce two auxiliary parameters. For a fixed scale sns_n, they consider coverings by sets whose diameters lie within the admissible range determined by θ\theta. Specifically, the aim is to minimise the quantity

−mlog⁡2log⁡sm\frac{-m\log2}{\log s_m}

where the index mm ranges over all the values satisfying sn1/θ≤sm≤sns_n^{1/\theta}\leq s_m\leq s_n; that is, over all admissible diameters sms_m that may be used to cover the set at scale sns_n, in accordance with the θ\theta-intermediate dimension constraints. Since 0<sm<10<s_m<1, we have

−mlog⁡2log⁡sm=log⁡2−log⁡(sm1/m),\frac{-m\log2}{\log s_m}=\frac{\log2}{-\log\left(s_m^{1/m}\right)},

so that minimising the left-hand side over the admissible range is equivalent to minimising sm1/ms_m^{1/m}. This motivates the following definition.

Definition 3.14. Let a={an}n∈Na=\{a_n\}_{n\in\mathbb{N}} be a positive, non-increasing, summable sequence, let θ∈(0,1]\theta\in(0,1] and let r∈Nr\in\mathbb{N}. We define γθ,a(r)\gamma_{\theta,a}(r) and ρθ,a(r)\rho_{\theta,a}(r) as follows:

γθ,a(r):=max⁡{m∈N:sm≥sr1/θ},\gamma_{\theta,a}(r):=\max\left\{m\in\mathbb{N}:s_m\geq s_r^{1/\theta}\right\},

and

ρθ,a(r):=arg⁡min⁡r≤m≤γθ,a(r)+1{sm1/m}.\rho_{\theta,a}(r):=\underset{r\leq m\leq\gamma_{\theta,a}(r)+1}{\arg\min}\left\{s_m^{1/m}\right\}.

The following results are due to Angelini and Molter [2]. Throughout, ρθ,a\rho_{\theta,a} is the parameter of Definition Definition 3.14, and we abbreviate

Λθ,n(a):=−ρθ,a(n)log⁡2log⁡sρθ,a(n).\Lambda_{\theta,n}(a):=\frac{-\rho_{\theta,a}(n)\log2}{\log s_{\rho_{\theta,a}(n)}}.

Theorem 3.15 (The two extremal sets). Let a={an}a=\{a_n\} be positive, non-increasing and summable.

(i) If lim⁡n→∞(sn/sn+1)1/n=1\lim_{n\to\infty}(s_n/s_{n+1})^{1/n}=1, then for every θ∈(0,1]\theta\in(0,1],

dim⁡‾θCa=lim sup⁡n→∞Λθ,n(a),dim⁡‾θCa=lim inf⁡n→∞Λθ,n(a)=dim⁡HCa=dim⁡BCa.\overline{\dim}_{\theta} C_a=\limsup_{n\to\infty}\Lambda_{\theta,n}(a),\qquad\underline{\dim}_{\theta} C_a=\liminf_{n\to\infty}\Lambda_{\theta,n}(a)=\dim_H C_a=\dim_B C_a.

(ii) For every θ∈[0,1]\theta\in[0,1],

dim⁡‾θDa=θdim⁡‾BCa1−(1−θ)dim⁡‾BCa,dim⁡‾θDa=θdim⁡‾BCa1−(1−θ)dim⁡‾BCa.\overline{\dim}_{\theta} D_a=\frac{\theta\overline{\dim}_B C_a}{1-(1-\theta)\overline{\dim}_B C_a},\qquad\underline{\dim}_{\theta}D_a=\frac{\theta\underline{\dim}_B C_a}{1-(1-\theta)\underline{\dim}_B C_a}.

Part (i) is a genuine formula rather than a mere bound: for CaC_a the lower intermediate dimension is constant in θ\theta and equal to the Hausdorff dimension, so the interpolation is degenerate from below, while the upper one varies. Part (ii) requires no hypothesis on aa beyond summability, and the value depends on aa only through the box dimension, which is the same for every set in Ca\mathcal{C}_a.

Theorem 3.16 (Extremality). Under the hypotheses of Theorem 3.15 (i), for every θ∈(0,1]\theta\in(0,1] and every E∈CaE \in\mathcal{C}_a,

dim⁡‾θDa≤dim⁡‾θE≤dim⁡‾θCa,dim⁡‾θDa≤dim⁡‾θE≤dim⁡‾θCa.\overline{\dim}_{\theta} D_a \le\overline{\dim}_{\theta} E \le\overline{\dim}_{\theta} C_a,\qquad\underline{\dim}_{\theta} D_a \le\underline{\dim}_{\theta} E \le\underline{\dim}_{\theta} C_a.

Thus CaC_a and DaD_a bracket the whole family, extending to every θ∈(0,1]\theta\in(0,1] the classical picture at θ=0\theta=0, where CaC_a maximises and DaD_a minimises the Hausdorff dimension. The lower bounds hold for every non-increasing aa, with no regularity assumption.

Definition 3.17. A positive, non-increasing, summable sequence a={an}a=\{a_n\} is level comparable if

2<inf⁡nsnsn+1≤sup⁡nsnsn+1<∞,2<\inf_n\frac{s_n}{s_{n+1}}\le\sup_n\frac{s_n}{s_{n+1}}<\infty,

equivalently, if there are constants 0<τ≤λ<1/20<\tau\le\lambda<1/2 with τ≤sn+1/sn≤λ\tau\le s_{n+1}/s_n\le\lambda for all nn.

The right-hand inequality is automatic whenever aa is doubling; the content of the definition is the left-hand one, namely that the ratios sn+1/sns_{n+1}/s_n stay bounded away from 1/21/2. For a central Cantor set it says exactly that the ratios of dissection are bounded away from 00 and from 1/21/2, and it gives sk≍a2ks_k\asymp a_{2k}.

Theorem 3.18 (Full range). Let θ∈(0,1]\theta\in(0,1] and let aa be level comparable. Then

{dim⁡θE:E∈Ca}=[dim⁡θDa,dim⁡θCa],\{\dim_{\theta} E:E\in\mathcal{C}_a\}=[\dim_{\theta}D_a,\dim_{\theta}C_a],
{dim⁡‾θE:E∈Ca}=[dim⁡‾θDa,dim⁡‾θCa].\{\underline{\dim}_{\theta} E:E\in\mathcal{C}_a\}=[\underline{\dim}_{\theta}D_a,\underline{\dim}_{\theta}C_a].

Every intermediate value is therefore attained: for fixed θ\theta and any ss in the corresponding interval there is a complementary set realising it exactly. Theorem 3.18 is the θ\theta-intermediate analogue of the Besicovitch–Taylor theorem, which is the case θ=0\theta=0.

Theorem 3.18 describes, for each fixed θ\theta, the set of values attained across Ca\mathcal{C}_a. A finer question is which functions of θ\theta arise. As recalled in § 2, Banaji and Rutar [7] characterised the functions that occur as the intermediate dimensions of a bounded subset of Rn\mathbb{R}^n; combining their result with [2] suggests the following: given a fixed sequence aa, which functions θ↦h(θ)\theta\mapsto h(\theta) arise as the intermediate dimension of a complementary set E∈CaE\in\mathcal{C}_a?

Some partial results in that direction were obtained in [2]. If aa is level comparable, then for every s>1s>1 there is a set E∈CaE\in\mathcal{C}_a with

dim⁡‾θE=max⁡{dim⁡‾θCas,θdim⁡‾BCa1−(1−θ)dim⁡BCa}=:ψs(θ).\underline{\dim}_{\theta}E=\max\left\{\frac{\overline{\dim}_{\theta}C_a}{s},\frac{\theta\overline{\dim}_{B}C_a}{1-(1-\theta)\dim_{B}C_a}\right\}=:\psi_s(\theta).

so each ψs\psi_s is realised in Ca\mathcal{C}_a. As s→1+s \to1^+ the functions ψs\psi_s increase to dim⁡‾θCa\overline{\dim}_\theta C_a, and as s→∞s \to\infty they decrease to dim⁡‾θDa\underline{\dim}_\theta D_a, so the family {ψs}s>1\{\psi_s\}_{s>1} sweeps a one-parameter curve joining the two extremal profiles of Theorem 3.16. On the other hand, every E∈CaE \in\mathcal{C}_a satisfies

θdim⁡‾BCa1−(1−θ)dim⁡‾BCa≤dim⁡‾θE≤dim⁡‾θCa.\frac{\theta\overline{\dim}_B C_a}{1-(1-\theta)\overline{\dim}_B C_a} \leq\overline{\dim}_\theta E \leq\overline{\dim}_\theta C_a.

A one-parameter family of realised profiles inside a two-sided bound is far from a characterisation, and the following remains open.

Question 3.19. Let aa be a level comparable sequence. Describe the structure of the families of functions

FU(a):={θ↦dim⁡‾θE:E∈Ca},FL(a):={θ↦dim⁡‾θE:E∈Ca},\mathcal{F}^{U}(a):=\left\{\theta\mapsto\overline{\dim}_\theta E:E\in\mathcal{C}_a\right\},\qquad\mathcal{F}^{L}(a):=\left\{\theta\mapsto\underline{\dim}_\theta E:E\in\mathcal{C}_a\right\},

that is, the collections of all functions arising, respectively, as the upper and lower θ\theta-intermediate dimensions of complementary sets associated with aa.

Assouad type dimensions

The Assouad dimension and the lower Assouad dimension of sets in Ca\mathcal{C}_a were studied in [23]. There, the authors prove that, given a non-increasing summable sequence a={an}a=\{a_n\},

dim⁡ACa=lim sup⁡n→∞(sup⁡knlog⁡2log⁡(sk/sk+n)),\dim_A C_a=\limsup_{n\to\infty}\left(\sup_k\frac{n\log2}{\log(s_k/s_{k+n})}\right),

and

dim⁡LCa=lim inf⁡n→∞(inf⁡knlog⁡2log⁡(sk/sk+n)).\dim_L C_a=\liminf_{n\to\infty}\left(\inf_k\frac{n\log2}{\log(s_k/s_{k+n})}\right).

Moreover, it was shown that dim⁡ADa=0\dim_A D_a=0 or 11, and if EE is any complementary set in Ca\mathcal{C}_a, then

dim⁡ACa≤dim⁡AE≤dim⁡ADa,\dim_A C_a\leq\dim_A E\leq\dim_A D_a,

and

0=dim⁡LDa≤dim⁡LE≤dim⁡LCa.0=\dim_L D_a\leq\dim_L E\leq\dim_L C_a.

Furthermore, if the sequence aa is doubling, then

{dim⁡AE:E∈Ca}=[dim⁡ACa,1],\{\dim_A E:E\in\mathcal{C}_a\}=[\dim_A C_a,1],

and if, instead of doubling, one assumes that inf⁡nsn/sn+1>2\inf_n s_n/s_{n+1}>2, then

{dim⁡LE:E∈Ca}=[0,dim⁡LCa].\{\dim_L E:E\in\mathcal{C}_a\}=[0,\dim_L C_a].

The Φ\Phi-dimensions introduced in § 2 give a one-parameter refinement of the Assouad-dimension results recalled above.

Theorem 3.20 ([24 Prop. 4.2, Thm. 4.3, Cor. 4.7]). Let aa be a positive, non-increasing, summable sequence and let Φ\Phi be an admissible function. Then, for every E∈CaE\in\mathcal{C}_a,

dim⁡‾ΦE≤dim⁡‾ΦDaanddim⁡‾ΦE≥dim⁡‾ΦDa=0.\overline{\dim}_\Phi E\leq\overline{\dim}_\Phi D_a\qquad\text{and}\qquad\underline{\dim}_\Phi E\geq\underline{\dim}_\Phi D_a=0.

If aa is level comparable, then in addition

dim⁡‾ΦE≥dim⁡‾ΦCaanddim⁡‾ΦE≤dim⁡‾ΦCa,\overline{\dim}_\Phi E\geq\overline{\dim}_\Phi C_a\qquad\text{and}\qquad\underline{\dim}_\Phi E\leq\underline{\dim}_\Phi C_a,

so that CaC_a and DaD_a are extremal in Ca\mathcal{C}_a for both dimensions. If moreover Φ(x)→p\Phi(x) \to p for some p∈[0,∞]p \in[0,\infty], every intermediate value is attained:

{dim⁡‾ΦE:E∈Ca}=[dim⁡‾ΦCa,dim⁡‾ΦDa],\{\overline{\dim}_{\Phi} E : E \in\mathcal{C}_a\} = [\overline{\dim}_{\Phi} C_a,\overline{\dim}_{\Phi} D_a],
{dim⁡‾ΦE:E∈Ca}=[0,dim⁡‾ΦCa].\{\underline{\dim}_{\Phi} E : E \in\mathcal{C}_a\} = [0,\underline{\dim}_{\Phi} C_a].

The same conclusions hold for the quasi-Assouad dimensions, with dim⁡‾ΦCa\overline{\dim}_{\Phi} C_a and dim⁡‾ΦCa\underline{\dim}_{\Phi} C_a replaced by dim⁡qACa\dim_{qA} C_a and dim⁡qLCa\dim_{qL} C_a.

For the lower Φ\Phi-dimension the pattern is the familiar one: the countable rearrangement DaD_a is minimal, with value 0 because it has isolated points, and the Cantor set CaC_a is maximal, exactly as for the Hausdorff and lower Assouad dimensions. For the upper Φ\Phi-dimension, however, the roles are reversed with respect to the Hausdorff dimension case: here CaC_a realises the smallest value in Ca\mathcal{C}_a and DaD_a the largest.

Remark 3.21. If aa is doubling and Φ(x)→0\Phi(x) \to0, then dim⁡‾BDa>0\overline{\dim}_{B}D_a > 0 and hence dim⁡‾ΦDa≥dim⁡qADa=1\overline{\dim}_{\Phi}D_a \geq\dim_{qA}D_a = 1, so the upper range in Theorem 3.20 is the full interval [dim⁡‾ΦCa,1][\overline{\dim}_{\Phi}C_a,1]. At the other extreme, if Φ(x)→∞\Phi(x) \to\infty then dim⁡ΦE=dim⁡BE\dim_{\Phi}E = \dim_{B}E, which is the same for every E∈CaE \in\mathcal{C}_a, and the interval degenerates to a point.

Theorem 3.20 describes the full range of values across Ca\mathcal{C}_a; one may instead ask for the typical value under a natural random model. Attach to each index n∈Nn \in\mathbb{N} an independent tag ωn\omega_n, uniformly distributed on [0,1][0,1], and order the gaps by increasing tag. This induces a random total order on N\mathbb{N} in which every ordering of a finite set of indices is equally likely and disjoint sets of indices are ordered independently, and hence a random element EωE_{\omega} of Ca\mathcal{C}_a. The following dichotomy identifies a sharp threshold, in the depth of the scales examined, separating two completely different almost sure behaviours.

Theorem 3.22 ([25]). Let aa be a level comparable sequence and set Ψ(x)=log⁡∣log⁡x∣∣log⁡x∣\Psi(x) = \frac{\log|\log x|}{|\log x|}. Then, almost surely:

(i) if Φ(x)≫Ψ(x)\Phi(x) \gg\Psi(x) as x→0x \to0, then dim⁡‾ΦEω=dim⁡‾ΦCa\overline{\dim}_{\Phi} E_{\omega} = \overline{\dim}_{\Phi} C_a and dim⁡‾ΦEω=dim⁡‾ΦCa\underline{\dim}_{\Phi} E_{\omega} = \underline{\dim}_{\Phi} C_a;

(ii) if Φ(x)≪Ψ(x)\Phi(x) \ll\Psi(x) as x→0x \to0, then dim⁡‾ΦEω=1\overline{\dim}_{\Phi} E_{\omega} = 1 and dim⁡‾ΦEω=0\underline{\dim}_{\Phi} E_{\omega} = 0.

Since the Assouad dimensions correspond to Φ≡0\Phi\equiv0, which falls in case (ii), and the quasi-Assouad dimensions arise from functions in case (i), Theorem 3.22 yields the following.

Corollary 3.23. Let aa be a level comparable sequence. Then, almost surely, E=EωE = E_{\omega} satisfies

dim⁡AE=1,dim⁡LE=0,\dim_A E = 1,\qquad\dim_L E = 0,

while

dim⁡qAE=dim⁡qACa,dim⁡qLE=dim⁡qLCa.\dim_{qA} E = \dim_{qA} C_a,\qquad\dim_{qL} E = \dim_{qL} C_a.

That is, the almost sure Assouad dimensions of a random rearrangement are always extremal, whatever CaC_a may be, whereas the almost sure quasi-Assouad dimensions always agree with those of CaC_a. In particular, since a set is uniformly perfect

precisely when its lower Assouad dimension is positive, almost no rearrangement is uniformly perfect. The critical regime Φ≍Ψ\Phi\asymp\Psi is not covered by Theorem 3.22.

Dynamically defined Cantor sets

The sections above study the whole family Ca\mathcal{C}_a of complementary sets through the combinatorial and measure-theoretic data of the defining sequence aa. From now on we restrict to the distinguished Cantor set CaC_a. A complementary viewpoint, developed in the second half of García’s PhD thesis [21], defines Cantor sets dynamically, as attractors of iterated function systems (IFS), and asks the converse question: which sequence-defined Cantor sets CaC_a arise this way, with what regularity? This dynamical viewpoint is then applied to a genuinely new question—the arithmetic sum Cp+CqC_p+C_q of two members of the “non-linear” family Cp:=C{1/np}C_p:=C_{\{1/n^p\}}—via the classical notion of thickness.

Iterated function systems and hyperbolic attractors

Definition 4.1 (IFS and attractor). Let D⊂RD\subset\mathbb{R} be closed and nonempty. A finite family {f0,…,fm}\{f_0,\ldots,f_m\} of maps fi:D→Df_i:D\to D is an iterated function system (IFS) if each fif_i is eventually a contraction: there are ri<1r_i<1 and n0≥1n_0\geq1 with

∣fin(x)−fin(y)∣≤ri∣x−y∣\left|f_i^n(x)-f_i^n(y)\right|\leq r_i\left|x-y\right|

for all x,y∈Dx,y\in D, n≥n0n\geq n_0.

Theorem 4.2 (Hutchinson [28]). Every IFS {f0,…,fm}\{f_0,\ldots,f_m\} has a unique nonempty compact attractor KK satisfying

K=⋃i=0mfi(K),K=\bigcup_{i=0}^{m}f_i(K),

and fn(A)→Kf^n(A)\to K in the Hausdorff metric on compacta, for every nonempty compact A⊂DA\subset D with fi(A)⊂Af_i(A)\subset A for all ii.

When the fif_i are similarities with ratios rir_i, the system is self-similar and KK is a self-similar set; the unique s≥0s\geq0 solving

∑iris=1\sum_i r_i^s=1

is the similarity dimension, and always Hs(K)≤∣K∣s<∞\mathcal{H}^s(K)\leq|K|^s<\infty, so dim⁡HK≤s\dim_{\mathrm H}K\leq s. Under the open set condition (an open bounded VV with ⋃ifi(V)⊂V\bigcup_i f_i(V)\subset V disjointly), Hutchinson showed Hs(K)>0\mathcal{H}^s(K)>0, so dim⁡HK=dim⁡BK=s\dim_{\mathrm H}K=\dim_{\mathrm B}K=s; Bandt–Graf and Schief later showed the open set condition is in fact equivalent to Hs(K)>0\mathcal{H}^s(K)>0—but not to the packing analogue, since Peres–Simon–Solomyak [36] exhibited self-similar sets on the line with Hs(K)=0\mathcal{H}^s(K)=0 yet 0<Ps(K)<∞0<\mathcal{P}^s(K)<\infty in their common dimension ss.

Definition 4.3 (Regular system). An IFS {f0,…,fm}\{f_0,\ldots,f_m\} on a closed interval II is regular if each fi∈C1(I)f_i\in C^1(I) and there are 0<α<1<1/β0<\alpha<1<1/\beta, c≥1c\geq1, b≥1b\geq1 with, for every nn and every word ω∈{0,…,m}n\omega\in\{0,\ldots,m\}^n: (contractivity) αn<∣fω′(x)∣<cβn\alpha^n<|f_\omega'(x)|<c\beta^n for all x∈Ix\in I; (bounded distortion) b−1≤∣fω′(x)∣/∣fω′(y)∣≤bb^{-1}\leq|f_\omega'(x)|/|f_\omega'(y)|\leq b for all x,y∈Ix,y\in I, where

fω:=fω1∘⋯∘fωn.f_\omega:=f_{\omega_1}\circ\cdots\circ f_{\omega_n}.

Systems with η\eta-Hölder continuous derivatives, or derivatives satisfying the Dini condition ∫01mfi′(t) dt/t<∞\int_0^1 m_{f_i}'(t)\,\mathrm{d}t/t<\infty (with mfim_{f_i} the modulus of continuity), automatically have bounded distortion. For a regular system, the sub-multiplicative quantity

∑ω∈{0,…,m}n∣fω′(x)∣s\sum_{\omega\in\{0,\ldots,m\}^n}|f_\omega'(x)|^s

grows like exp⁡(nΥ(s))\exp(n\Upsilon(s)) for a function Υ\Upsilon independent of xx; Υ\Upsilon is continuous, strictly decreasing, with Υ(−∞)=+∞\Upsilon(-\infty)=+\infty, Υ(+∞)=−∞\Upsilon(+\infty)=-\infty (so it has a unique zero), and equals log⁡∑iris\log\sum_i r_i^s in the self-similar case. This is Bowen’s equation Υ(s)=0\Upsilon(s)=0, originally introduced to compute the Hausdorff dimension of quasicircles [12].

Theorem 4.4 ([21]). If KK is the attractor of a regular system and Υ(s)=0\Upsilon(s)=0, then P0s(K)<∞\mathcal{P}^{s}_{0}(K)<\infty—unconditionally, with no separation hypothesis on the system.

Theorem 4.5 ([21 Theorem 3.23]). If a regular system satisfies the open set condition, then 0<Hs(K)0<\mathcal{H}^{s}(K) where Υ(s)=0\Upsilon(s)=0; in particular s=dim⁡HK=dim⁡BKs=\dim_{H}K=\dim_{B}K.

Theorems 4.4 and 4.5 together give 0<Hs(K)≤P0s(K)<∞0<\mathcal{H}^{s}(K)\leq\mathcal{P}^{s}_{0}(K)<\infty under the open set condition: the attractor of a regular system is an ss-set, and the same exponent regularises the covering and the packing process at once. This is the dynamical counterpart of Theorem 3.3, where the gauge hah_a played the role now played by ss—with the difference that there a general dimension function was needed, whereas here regularity of the system forces a power to suffice.

Definition 4.6 (Cookie-cutter set). A regular two-map system {f0,f1}\{f_0,f_1\} on [0,1][0,1] is cookie-cutter (or hyperbolic) if it satisfies the strong separation condition

0=f0(0)<f0(1)<f1(0)<f1(1)=10=f_0(0)<f_0(1)<f_1(0)<f_1(1)=1

together with a strengthening of bounded distortion (uniform convergence of local distortion ratios to 11 at small scales); the attractor is a cookie-cutter (or hyperbolic) Cantor set. If the defining derivatives are η\eta-Hölder continuous, the set is C1+ηC^{1+\eta}-hyperbolic.

Theorem 4.7 ([21]). If CC is the attractor of a cookie-cutter system and Υ(s)=0\Upsilon(s)=0, then CC is a Cantor set (compact, perfect, totally disconnected) with 0<Hs(C)≤P0s(C)<∞0<\mathcal{H}^{s}(C)\leq\mathcal{P}^{s}_{0}(C)<\infty, and L(C)=0\mathcal{L}(C)=0.

Strong separation also gives CC a natural symbolic coding by Ω+={0,1}N\Omega^{+}=\{0,1\}^{\mathbb{N}} that is order-preserving for the lexicographic order on Ω+\Omega^{+} (Bedford [8]), which will be used below to encode the scale function.

The family CpC_p as a dynamically defined Cantor set

Recall from Example 3.2 that Cp=CaC_p=C_a for an=n−pa_n=n^{-p} with p>1p>1, that dim⁡HCp=1/p\dim_{H}C_p=1/p, and that CpC_p is a 1/p1/p-set. The main result of this subsection realises CpC_p as the attractor of a cookie-cutter system, and identifies the optimal smoothness of such a realisation.

Theorem 4.8 ([22 Theorem 1]). The set CpC_p is the attractor of a cookie-cutter system {fp,0,fp,1}\{f_{p,0},f_{p,1}\} whose derivatives are 1/p1/p-Hölder continuous, so CpC_p is C1+1/pC^{1+1/p}-hyperbolic. Moreover, 1/p1/p is the largest Hölder exponent attainable: no hyperbolic system respecting the step-kk interval structure of CpC_p has attractor CpC_p with derivatives of Hölder exponent greater than 1/p1/p. In this sense CpC_p is strictly C1+1/pC^{1+1/p}-hyperbolic.

Combined with Theorem 4.7, this recovers from the dynamical side the fact recalled above that CpC_p is a 1/p1/p-set: hyperbolicity forces 0<Hs(Cp)≤P0s(Cp)<∞0<\mathcal{H}^{s}(C_p)\leq\mathcal{P}^{s}_{0}(C_p)<\infty with s=1/ps=1/p.

This raises the converse question for central Cantor sets CaC_a, i.e. a2n+j=λna_{2^n+j}=\lambda_n, 0≤j<2n0\leq j<2^n: which decay conditions on {λn}\{\lambda_n\} make CaC_a a dynamically defined set, and with what smoothness? Write bn=xn/nb_n=x_n/n as usual and suppose aa satisfies the (Besicovitch–Taylor-type) decay condition

lim⁡k→∞2kb2ks=A,0<A<∞,\lim_{k\to\infty}2^k b_{2^k}^s=A,\qquad0<A<\infty,

which is strictly stronger than merely 0<lim inf⁡nbns≤lim sup⁡nbns<∞0<\liminf nb_n^s\le\limsup nb_n^s<\infty (i.e. 0<Hs(Ca)≤P0s(Ca)<∞0<\mathcal{H}^s(C_a)\le\mathcal{P}_0^s(C_a)<\infty, Theorem 3.5) and is genuinely needed: there are non-increasing sequences satisfying the latter but for which no continuously differentiable system has attractor CaC_a.

Theorem 4.9 ([21]). If aa is central and satisfies the decay condition above, then CaC_a is the attractor of a continuously differentiable system {f0,f1}\{f_0,f_1\} (with f0f_0, f1f_1 translates of each other) whose derivative equals the constant 2−1/s2^{-1/s} at every point of CaC_a itself.

The smoothness of this system away from CaC_a is governed by how fast the derivative’s oscillation on the gaps decays. Writing uk:=λk+1/λk−2−1/su_k:=\lambda_{k+1}/\lambda_k-2^{-1/s} for the discrepancy at step kk:

Theorem 4.10 ([21]). With aa, ss as in Theorem 4.9: if {uk}∈ℓ1\{u_k\}\in\ell^1, the system has bounded distortion; conversely, if the system’s derivatives satisfy the Dini condition, then {uk}∈ℓ1\{u_k\}\in\ell^1. Moreover, the system is strictly C1+ηC^{1+\eta}-hyperbolic if and only if uk∼λkηu_k\sim\lambda_k^\eta.

Theorem 4.10 is a complete dictionary between the decay rate of the sequence and the Hölder smoothness of the dynamically-realising system, and it recovers Theorem 4.8 for CpC_p as a special case (with η=1/p\eta=1/p). A characterisation of central sets admitting Hölder-continuously-differentiable realising systems, by different methods, was independently obtained by Bamón–Moreira–Plaza–Vera [4].

Definition 4.11 (Sullivan’s scale function). Let CC be a C1+ηC^{1+\eta}-hyperbolic Cantor set with symbolic coding Ω+\Omega^+; let Ω−\Omega^- denote the analogous coding of the “dual” Cantor set obtained by composing along the opposite (past) direction. For α∈Ω−\alpha\in\Omega^-, let Rn(α)∈R3R_n(\alpha)\in\mathbb{R}^3 record the ratios (left interval, gap, right interval)/(total) at step nn around α\alpha; these converge, as n→∞n\to\infty, to a limit R(α)R(\alpha) that is η\eta-Hölder continuous. The map R:Ω−→R3R:\Omega^-\to\mathbb{R}^3 is the scale function of CC.

Theorem 4.12 (Sullivan [41], proof in [9]). Two C1+ηC^{1+\eta}-hyperbolic Cantor sets are C1+ηC^{1+\eta}-conjugate if and only if they have the same scale function.

That is, the scale function is a complete C1C^1-invariant for hyperbolic Cantor sets—it “remembers” the fine geometry of the set at all small scales, and two hyperbolic sets are smoothly equivalent exactly when this invariant coincides.

Theorem 4.13 ([22 Theorem 2]). (Cp,{fp,0,fp,1})(C_p,\{f_{p,0},f_{p,1}\}) is C1+1/pC^{1+1/p}-conjugate to the self-similar system

(Λ2−p,{2−px,  2−px+(1−2−p)}),\left(\Lambda_{2^{-p}},\left\{2^{-p}x,\;2^{-p}x+(1-2^{-p})\right\}\right),

via an order-preserving diffeomorphism hp:[0,1]→Iph_p:[0,1]\to I_p with 1/p1/p-Hölder derivative.

Theorem 4.13 follows from Theorem 4.12 by an explicit computation showing the two sets share the same (constant) scale function. It transports the self-similar (natural) measure μ2−p\mu_{2^{-p}} on Λ2−p\Lambda_{2^{-p}}—the classical Bernoulli convolution associated with parameter 2−p2^{-p}—to the invariant measure υp:=μ2−p∘hp−1\upsilon_p := \mu_{2^{-p}} \circ h_p^{-1} on CpC_p, which one shows is comparable to \mathcal{H}^{1/p}\mathbin{\vrule height 1.4ex depth -0.3ex width 0.07ex\vrule height 0.07ex depth -0.02ex width 0.7ex}C_p, since hph_p is bi-Lipschitz and Hausdorff measures change by at most bounded factors under bi-Lipschitz maps; we write \upsilon_p \approx\mathcal{H}^{1/p}\mathbin{\vrule height 1.4ex depth -0.3ex width 0.07ex\vrule height 0.07ex depth -0.02ex width 0.7ex}C_p. This equivalence is the bridge used in the next subsection to transfer results about Bernoulli convolutions to the non-linear family {Cp}\{C_p\}.

Sums of Cantor sets and thickness

Motivation: Bernoulli convolutions

The self-similar measure μr\mu_r on Λr\Lambda_r (with weights {1/2,1/2}\{1/2,1/2\}) is the classical Bernoulli convolution: the law of (1−r)∑n≥0ϵnrn(1-r)\sum_{n\geq0}\epsilon_n r^n for i.i.d. fair coin flips ϵn∈{0,1}\epsilon_n \in\{0,1\}. Every self-similar measure is of pure type (singular or absolutely continuous); for μr\mu_r this dichotomy has been studied intensively since the 1930s.

Theorem 5.1 (Erdős, 1939 [16]). If r∈(1/2,1)r \in(1/2,1) and 1/r1/r is a Pisot number (an algebraic integer >1>1 all of whose Galois conjugates have modulus <1<1), then μr\mu_r is singular; in fact μr^(ξ)↛0\widehat{\mu_r}(\xi) \not\to0 as ξ→∞\xi\to\infty.

Whether the set S⊥S_\perp of r∈(1/2,1)r \in(1/2,1) with μr\mu_r singular contains anything beyond reciprocals of Pisot numbers is open; on the positive side, Solomyak [40] showed μr\mu_r is absolutely continuous for a.e. r∈(1/2,1)r \in(1/2,1), and Shmerkin [39] showed dim⁡H(S⊥)=0\dim_H(S_\perp)=0.

Thickness and Newhouse's Gap Lemma

A general bound on the dimension of a sum follows from the Lipschitz map (x,y)↦x+y(x,y)\mapsto x+y: if dim⁡HF=dim⁡BF\dim_H F=\dim_B F then dim⁡H(E+F)≤dim⁡HE+dim⁡HF\dim_H(E+F)\leq\dim_H E+\dim_H F (via the product formula for box-regular factors); since dim⁡HCq=dim⁡BCq\dim_H C_q=\dim_B C_q, this gives L(Cp+Cq)=0\mathcal{L}(C_p+C_q)=0 whenever 1/p+1/q<11/p+1/q<1.

For the complementary question—when does a sum contain an interval—the classical tool is Newhouse’s notion of thickness, central to the study of homoclinic bifurcations [34].

Definition 5.2 (Thickness). Let CC be a Cantor set on the line with a gap LL. A bridge of LL is a maximal interval sharing an endpoint with LL that meets no gap at least as long as LL. The thickness of CC is

τ(C):=inf⁡{∣B∣∣L∣:(B,L) is a bridge/gap pair of C}.\tau(C):=\inf\left\{\frac{|B|}{|L|}:(B,L)\text{ is a bridge/gap pair of }C\right\}.

Thickness is invariant under translation and scaling: τ(C+t)=τ(sC)=τ(C)\tau(C+t)=\tau(sC)=\tau(C).

Lemma 5.3 (Newhouse’s Gap Lemma [32]). If C1,C2C_1,C_2 are Cantor sets on the line with τ(C1)⋅τ(C2)≥1\tau(C_1)\cdot\tau(C_2)\geq1, then either C1C_1 is contained in a gap of C2C_2, or C2C_2 is contained in a gap of C1C_1, or C1∩C2≠∅C_1\cap C_2\neq\varnothing.

Proposition 5.4 ([34]). If τ(C1)⋅τ(C2)≥1\tau(C_1)\cdot\tau(C_2)\geq1, then C1+C2C_1+C_2 is a finite union of closed intervals (so contains an open interval); if in addition no translate of either set lies in a bounded gap of the other, C1+C2C_1+C_2 is a single interval.

For example, τ(Λr)=r/(1−2r)\tau(\Lambda_r)=r/(1-2r), so Λr+Λr=[0,2]\Lambda_r+\Lambda_r=[0,2] once r≥1/3r\ge1/3. For CpC_p, a direct computation with the step-interval lengths gives the two-sided bound

12p⋅12p−2≤τ(Cp)≤(23)p12p−2,\frac{1}{2p}\cdot\frac{1}{2^p-2}\leq\tau(C_p)\leq\left(\frac{2}{3}\right)^p\frac{1}{2^p-2},

which, via Proposition 5.4, only yields that Cp+CqC_p+C_q contains an interval under the condition 12p(2p−2)⋅12q(2q−2)≥1\frac{1}{2^p(2^p-2)}\cdot\frac{1}{2^q(2^q-2)}\geq1—much more restrictive than the condition 12p−2⋅12q−2≥1\frac{1}{2^p-2}\cdot\frac{1}{2^q-2}\geq1 that one would expect by analogy with the self-similar sets Λ2−p\Lambda_{2-p} and Λ2−q\Lambda_{2-q}.

The gap can be closed using local thickness τloc(C,x):=lim⁡k→∞τ(C∩Ixk)\tau_{\mathrm{loc}}(C,x):=\lim_{k\to\infty}\tau(C\cap I_x^k) (the thickness of the sub-Cantor-sets shrinking to xx), for which the analogous gap-lemma argument only requires the local thickness product to exceed 1 pointwise:

Theorem 5.5 ([22]). The local thickness of CpC_p is constant, equal to τloc(Cp)=1/(2p−2)\tau_{\mathrm{loc}}(C_p)=1/(2^p-2). Consequently, if 12p−2⋅12q−2>1\frac{1}{2^p-2}\cdot\frac{1}{2^q-2}>1, then Cp+CqC_p+C_q contains an open interval—matching, up to the boundary case, the bound one would guess from the conjugate self-similar sets Λ2−p\Lambda_{2-p}.

Global (almost-everywhere) results

Beyond the thickness threshold, one can ask for the exact Hausdorff dimension or positivity of Lebesgue measure of Cp+CqC_p+C_q across the full range of parameters. Adapting the transversality method of Peres–Solomyak [37] for homogeneous self-similar families—using the C1+1/pC^{1+1/p} conjugacy of Theorem 4.13 to control the derivative, in pp, of the projection Γp:Ω+→Cp\Gamma_p:\Omega^+\to C_p—gives:

Theorem 5.6 ([22]). For every compact K⊂RK\subset\mathbb{R}: (a) for a.e. p>1p>1 with dim⁡HK+1/p<1\dim_H K+1/p<1, dim⁡H(K+Cp)=dim⁡HK+1/p\dim_H(K+C_p)=\dim_H K+1/p; (b) for a.e. p>1p>1 with dim⁡HK+1/p>1\dim_H K+1/p>1, L(K+Cp)>0\mathcal{L}(K+C_p)>0.

Corollary 5.7 ([22]). For fixed q>1q>1: for a.e. p>1p>1, dim⁡H(Cp+Cq)=min⁡{1/p+1/q,1}\dim_H(C_p+C_q)=\min\{1/p+1/q,1\}.

Corollary 5.7 is the exact non-linear analogue, for the family {Cp}\{C_p\}, of the Peres–Schmerkin formula dim⁡H(Λr+Λs)=min⁡{dim⁡HΛr+dim⁡HΛs,1}\dim_H(\Lambda_r+\Lambda_s)=\min\{\dim_H\Lambda_r+\dim_H\Lambda_s,1\} (valid whenever log⁡s/log⁡r\log s/\log r is irrational [35]), obtained here for a full-measure set of the parameter pp rather than under an arithmetic condition—reflecting the extra transversality available from the non-linearity of the conjugacy hph_p.

The absolute continuity half of the Peres–Solomyak method also transfers, via a Frostman-measure argument, to the invariant measures υp\upsilon_p:

Theorem 5.8 ([22]). Fix q>1q>1. For a.e. p>1p>1 with 1/p+1/q>11/p+1/q>1, the convolution υq∗υp\upsilon_q*\upsilon_p is absolutely continuous with density in L2(R)L^2(\mathbb{R}); equivalently (since \upsilon_p\approx\mathcal{H}^{1/p}\mathbin{\vrule height 1.4ex depth -0.3ex width 0.07ex\vrule height 0.07ex depth -0.02ex width 0.7ex}C_p), the convolution \mathcal{H}^{1/q}\mathbin{\vrule height 1.4ex depth -0.3ex width 0.07ex\vrule height 0.07ex depth -0.02ex width 0.7ex}C_q*\mathcal{H}^{1/p}\mathbin{\vrule height 1.4ex depth -0.3ex width 0.07ex\vrule height 0.07ex depth -0.02ex width 0.7ex}C_p is absolutely continuous with L2L^2 density.

Remark 5.9. The condition “for a.e. pp, qq fixed” in Theorem 5.8 and Corollary 5.7 cannot currently be upgraded to “for every p,qp,q” or symmetrised in both parameters simultaneously; this is a limitation of the transversality method inherited from [37], already pointed out in [22]. Whether the Fourier transform υp^(ξ)\widehat{\upsilon_p}(\xi) decays to 0 as ξ→∞\xi\to\infty—the natural non-linear analogue of the Erdős/Salem theory of § 5 above—is entirely open: the non-linearity of the conjugacy hph_p destroys the product-of-cosines formula that makes μ^r\hat{\mu}_r tractable in the self-similar case.

The results of this section connect to a broader circle of ideas on sums of dynamically defined Cantor sets: Palis conjectured that the sum of two C1+ϵC^{1+\epsilon}-hyperbolic Cantor sets either has zero Lebesgue measure or contains an interval; this is false in general—Sannami [38] constructed a C∞C^\infty Cantor set AA such that A−AA-A has positive Lebesgue measure but empty interior, but true generically, in the sense of Moreira–Yoccoz [31], who showed that pairs (C1,C2)(C_1,C_2) of hyperbolic sets with dim⁡HC1+dim⁡HC2>1\dim_H C_1+\dim_H C_2>1 generically admit a stable intersection (a robust transversal intersection of C1C_1 and a translate of −C2-C_2), forcing C1+C2C_1+C_2 to contain an interval. Whether Λr+Λs\Lambda_r+\Lambda_s has nonempty interior throughout the range 1<dim⁡HΛr+dim⁡HΛs<1+δ1<\dim_H\Lambda_r+\dim_H\Lambda_s<1+\delta for small δ\delta remains open.

Open Problem 5.10 ([21]). Extend the transversality analysis above to the bi-parametric family {Cp,q}p>1,q∈R\{C_{p,q}\}_{p>1,q\in\mathbb{R}} (sequences an=(log⁡n)q/npa_n=(\log n)^q/n^p), using the lexicographic dimension dim⁡F2\dim_{\mathcal{F}_2} induced by F2:={hp,q}∪{x}∪{1}\mathcal{F}_2:=\{h_{p,q}\}\cup\{x\}\cup\{1\} in place of dim⁡H\dim_H; conjecturally, for fixed q,q′q,q' and a.e. (p,p′)(p,p') (or fixed qq and a.e. pp), dim⁡F2(Cp,q+Cp′,q′)=min⁡{hp,q⋅hp′,q′,x}\dim_{\mathcal{F}_2}(C_{p,q}+C_{p',q'})=\min\{h_{p,q}\cdot h_{p',q'},x\} in the sense of the order on F2\mathcal{F}_2. This would require bi-parametric analogues of Frostman’s lemma and of β\beta-energy adapted to dim⁡F2\dim_{\mathcal{F}_2}.

Concluding remarks

A single pattern runs through the results collected here. For every notion of size considered, the two canonical members of Ca\mathcal{C}_a are extremal, exactly or up to constants, and in most cases where a complete description of the attained values exists it is a closed interval with endpoints at those two sets; the packing measures, with their all-or-nothing dichotomy, are the exception. What changes from one notion to the next is which of them sits at which end.

For the Hausdorff dimension, and more generally for the Hausdorff hh-measures, the maximum is at CaC_a and the minimum at the countable set DaD_a; the same holds for the intermediate dimensions at every θ∈(0,1]\theta\in(0,1], and for the lower Assouad and lower Φ\Phi-dimensions. For the upper Assouad and upper Φ\Phi-dimensions the order is reversed, and so it is for the packing hh-premeasure: CaC_a is simultaneously the maximiser of Hh\mathcal{H}^h and the minimiser of P0h\mathcal{P}^h_0 within the family. The reversal is not an accident of proof technique. Covering-based and global quantities reward the even distribution of gaps that the Cantor construction produces, while packing-based and local ones reward the concentration that the decreasing rearrangement produces, and CaC_a and DaD_a are the extreme cases of each.

Against this variability stands the rigidity of the box dimensions, which take the same value on every member of Ca\mathcal{C}_a. The intermediate dimensions measure exactly how far a given notion sits from that rigid end: they run from the full-range behaviour of dim⁡H\dim_H at θ=0\theta=0 to the constancy of dim⁡B\dim_B at θ=1\theta=1, and Theorem 3.18 shows that for each θ\theta every value between the two extremes is attained, the interval collapsing to a point only at θ=1\theta=1.

Several questions remain open. The most substantial, stated as Question 3.19, asks not which values but which functions of θ\theta arise within a fixed family: the ambient answer for arbitrary compact sets is known from [7], and the profiles ψs\psi_s of § 3 give a one-parameter family of realised examples, but between these lies a genuine gap. For the Φ\Phi-dimensions, the almost sure behaviour in the critical regime Φ≍log⁡∣log⁡x∣/∣log⁡x∣\Phi\asymp\log\lvert\log x\rvert/\lvert\log x\rvert is not covered by the dichotomy of Theorem 3.22. In the Hausdorff–packing duality, the sharp constants and the extremal sets are known on the Hausdorff side only for middle-τ\tau Cantor sets, and on the packing side not even for the middle-third Cantor set. Finally, on the dynamical side, the almost-everywhere results for Cp+CqC_p+C_q cannot at present be made uniform in both parameters, and the decay of υcp\upsilon_{cp}—the non-linear counterpart of the Erdős–Salem theory for Bernoulli convolutions—is entirely open.

Acknowledgments

These notes grew out of three lectures given by the authors at the Instituto de Matemática Aplicada San Luis (Consejo Nacional de Investigaciones Científicas y Técnicas - Universidad Nacional de San Luis), San Luis, Argentina, in April 2026. We are very grateful to Sergio Favier for the invitation, and its warm hospitality. We also thank the participants for their questions and comments, which helped shape this text. (N. Angelini) UNSL–IMASL, CONICET, San Luis, Argentina Email address: [email protected]

(C. Cabrelli) Departamento de Matemática, FCEyN, Universidad de Buenos Aires, and IMAS–CONICET, Buenos Aires, Argentina Email address: [email protected]

(U. Molter) Departamento de Matemática, FCEyN, Universidad de Buenos Aires, and IMAS–CONICET, Buenos Aires, Argentina Email address: [email protected]

References

  1. [1]N. Angelini and U. Molter, Critical values for intermediate and box dimensions of projections and other images of compact sets, J. Fractal Geom. 12 (2025), no. 3–4, 369–394.DOI
  2. [2]N. Angelini and U. Molter, Intermediate dimensions and their applications, Proc. Roy. Soc. Edinburgh Sect. A, published online 2026, 1–19, doi:10.1017/prm.2026.10162.
  3. [3]N. E. Angelini, U. M. Molter, J. M. Tejada. Intermediate dimensions of measures: interpolating between Hausdorff and Minkowski dimensions. J. Math. Anal. Appl. 555 (2026), no. 1, Paper No. 130039.
  4. [4]R. Bamón, C. G. Moreira, S. Plaza, J. Vera. Differentiable structures of central Cantor sets. Ergodic Theory Dynam. Systems 17 (1997), no. 5, 1027–1042.
  5. [5]A. Banaji. Generalised intermediate dimensions. Monatsh. Math. 202 (2023), no. 3, 465–506.
  6. [6]A. Banaji and I. Kolossváry, Intermediate dimensions of Bedford–McMullen carpets with applications to Lipschitz equivalence, Adv. Math. 449 (2024), Paper No. 109735, 69 pp.
  7. [7]A. Banaji and A. Rutar, Attainable forms of intermediate dimensions, Ann. Fenn. Math. 47 (2022), no. 2, 939–960.
  8. [8]T. Bedford. Applications of dynamical systems theory to fractals—a study of cookie-cutter Cantor sets. In Fractal Geometry and Analysis (Montreal, 1989), NATO ASI Series C, vol. 346, Kluwer, 1991, 1–44.DOI
  9. [9]T. Bedford, A. M. Fisher. Ratio geometry, rigidity and the scenery process for hyperbolic Cantor sets. Ergodic Theory Dynam. Systems 17 (1997), no. 3, 531–564.arxiv.org/abs/math/9405217
  10. [10]A. S. Besicovitch, S. J. Taylor. On the complementary intervals of a linear closed set of zero Lebesgue measure. J. London Math. Soc. 29 (1954), 449–459.DOI
  11. [11]E. Best. A closed dimensionless linear set. Proc. Edinburgh Math. Soc. (2) 6 (1939), 105–108.DOI
  12. [12]R. Bowen. Hausdorff dimension of quasicircles. Inst. Hautes Études Sci. Publ. Math. 50 (1979), 11–25.
  13. [13]C. Cabrelli, U. B. Darji and U. M. Molter, Visible and invisible Cantor sets, in: Excursions in Harmonic Analysis, Vol. 2, 2013.arxiv.org/abs/1109.1174
  14. [14]C. Cabrelli, F. Mendivil, U. Molter, R. Shonkwiler. On the Hausdorff h-measure of Cantor sets. Pacific J. Math. 217 (2004), no. 1, 45–59.
  15. [15]R. O. Davies, Sets which are null or non-sigma-finite for every translation invariant measure, Mathematika 18 (1971), no. 2, 161–162.DOI
  16. [16]P. Erdős. On a family of symmetric Bernoulli convolutions. Amer. J. Math. 61 (1939), 974–976.DOI
  17. [17]K. J. Falconer. Techniques in Fractal Geometry. John Wiley & Sons, Chichester, 1997.
  18. [18]K. Falconer. Fractal Geometry: Mathematical Foundations and Applications, 3rd ed. Wiley, 2014.
  19. [19]K. Falconer, J. M. Fraser, T. Kempton. Intermediate dimensions. Math. Z. 296 (2020), 813–830.
  20. [20]J. M. Fraser, Assouad Dimension and Fractal Geometry, Cambridge Tracts in Mathematics, vol. 222, Cambridge University Press, Cambridge, 2020.
  21. [21]I. A. Garcia. Análisis fractal de conjuntos de Cantor no lineales. Tesis Doctoral, Universidad Nacional del Litoral, 2008.
  22. [22]I. Garcia, A family of smooth Cantor sets, Ann. Acad. Sci. Fenn. Math. 36 (2011), 21–45.
  23. [23]I. García, K. Hare, F. Mendivil. Assouad dimensions of complementary sets. Proc. Roy. Soc. Edinburgh Sect. A 150 (2020), 2585–2598.
  24. [24]I. García, K. E. Hare, and F. Mendivil, Intermediate Assouad-like dimensions, J. Fractal Geom. 8 (2021), no. 3, 201–245.
  25. [25]I. García, K. E. Hare, and F. Mendivil, Almost sure Assouad-like dimensions of complementary sets, Math. Z. 298 (2021), no. 3–4, 1201–1220.arxiv.org/abs/1903.07800
  26. [26]I. Garcia, U. Molter, R. Scotto. Dimension functions of Cantor sets. Proc. Amer. Math. Soc. 135 (2007), no. 10, 3151–3161.
  27. [27]K. E. Hare, F. Mendivil and L. Zuberman, The sizes of rearrangements of Cantor sets, Canad. Math. Bull. 56 (2013), no. 2, 354–365.
  28. [28]J. E. Hutchinson. Fractals and self-similarity. Indiana Univ. Math. J. 30 (1981), no. 5, 713–747.
  29. [29]F. Liu and L.-F. Xi, Quasi-Assouad dimension of fractals, J. Fractal Geom. 3 (2016), no. 2, 187–215.
  30. [30]C. G. Moreira, H. Erazo, and N. Angelini, Generalized Hausdorff dimension of irrationals with Lagrange value exactly 3, preprint, 2025, arXiv:2511.10328.arxiv.org/abs/2511.10328
  31. [31]C. G. T. de A. Moreira, J.-C. Yoccoz. Stable intersections of regular Cantor sets with large Hausdorff dimensions. Ann. of Math. (2) 154 (2001), no. 1, 45–96.
  32. [32]S. E. Newhouse, The abundance of wild hyperbolic sets and non-smooth stable sets for diffeomorphisms, Publ. Math. Inst. Hautes Études Sci. 50 (1979), 101–151.DOI
  33. [33]L. Olsen and D. L. Renfro, On the exact Hausdorff dimension of the set of Liouville numbers. II, Manuscripta Math. 119 (2006), no. 2, 217–224.DOI
  34. [34]J. Palis, F. Takens. Hyperbolicity and Sensitive Chaotic Dynamics at Homoclinic Bifurcations. Cambridge Studies in Advanced Mathematics, vol. 35, Cambridge University Press, 1993.
  35. [35]Y. Peres and P. Shmerkin, Resonance between Cantor sets, Ergodic Theory Dynam. Systems 29 (2009), no. 1, 201–221.
  36. [36]Y. Peres, K. Simon and B. Solomyak, Self-similar sets of zero Hausdorff measure and positive packing measure, Israel J. Math. 117 (2000), 353–379.DOI
  37. [37]Y. Peres, B. Solomyak. Self-similar measures and intersections of Cantor sets. Trans. Amer. Math. Soc. 350 (1998), no. 10, 4065–4087.DOI
  38. [38]A. Sannami, An example of a regular Cantor set whose difference set is a Cantor set with positive measure, Hokkaido Math. J. 21 (1992), no. 1, 7–24.
  39. [39]P. Shmerkin, On the exceptional set for absolute continuity of Bernoulli convolutions, Geom. Funct. Anal. 24 (2014), no. 3, 946–958.arxiv.org/abs/1303.3992
  40. [40]B. Solomyak. On the random series ∑ ±λ^n (an Erdős problem). Ann. of Math. (2) 142 (1995), no. 3, 611–625.DOI
  41. [41]D. Sullivan. Differentiable structures on fractal-like sets, determined by intrinsic scaling functions on dual Cantor sets. In The Mathematical Heritage of Hermann Weyl (Durham, NC, 1987), Proc. Sympos. Pure Math., vol. 48, Amer. Math. Soc., 1988, 15–23.DOI
  42. [42]Claude Tricot. Curves and fractal dimension. Springer Science & Business Media, 1995.DOI

Paper details

vlnc:W_efekeqj4
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 1

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents