betaFeedback

WorksMathematics

Stability of Carathéodory Outer Measures Generated by Spatial Gauges

Abstract

We study outer measures generated by covering costs that depend on both the diameter and the location of a covering set. After establishing the metric and Borel-regularity properties of this construction, we investigate its stability under perturbations of the spatial gauge. For power gauges, we prove a uniform logarithmic criterion: if the exponent perturbation multiplied by log⁡(1/r)\log(1/r) converges uniformly to λλ, then the induced measure is multiplied by e−λe^{-λ}. Pointwise asymptotic equivalence, however, is insufficient: on [0,1][0,1] we construct bounded gauges whose ratio is eventually equal to one at every fixed point, although their induced measures differ by a prescribed factor. In the converse direction, continuity of the ratio in the spatial variable on a Baire space yields equality of the measures on every compact subset of a suitable dense GδG_δ set. The results identify a concrete obstruction to pointwise-to-measure stability and a localization principle, while leaving open the corresponding question for all Borel sets.

Introduction

Hausdorff-type measures are constructed by assigning a cost to each member of a sufficiently fine covering and then minimizing the total cost. This procedure is a form of Carathéodory’s Method II [1, 2]; its classical theory is developed in, among other sources, [6, 10, 8, 7, 14, 5]. In the standard Hausdorff construction, the cost depends only on the diameter of a covering set. In several geometric settings, however, the cost also depends on the position of that set. Such spatial dependence appears in multifractal constructions [9] and in the theory of centred Hausdorff measures [11, 12, 13].

The purpose of this paper is to examine how the outer measure changes when the spatial gauge is perturbed, with particular emphasis on the distinction between uniform and pointwise asymptotic control. Let (X,d)(X,d) be a metric space and let h:X×(0,∞)→[0,∞]h:X\times(0,\infty)\to[0,\infty]. For a set A⊆XA\subseteq X of positive finite diameter, we assign the cost

ch(A)=inf⁡x∈Ah(x,diam⁡A),c_h(A)=\inf_{x\in A} h(x,\operatorname{diam} A),

and assign cost zero to the empty set and to sets of diameter zero. The corresponding δ\delta-contents are obtained by taking the infimum of the total cost over all countable δ\delta-coverings; their limit as δ↓0\delta\downarrow0 is denoted by Hh\mathcal{H}^{h}. The precise definitions are given in § 2.

Some basic facts about this construction are needed before stability can be discussed. We prove that each fixed-scale content is an outer measure and that the limiting set function is a metric outer measure. We also prove Borel regularity, with an explicit FσδF_{\sigma\delta} hull for every set. These statements provide the framework for the later arguments; the metric outer-measure principle and Carathéodory’s criterion themselves are classical, and we do not present them as the principal source of novelty.

The main stability results are as follows.

  1. Uniform comparison and logarithmic perturbations. Uniform asymptotic equivalence of two spatial gauges implies equality of their limiting outer measures. For gauges of the form

h(x,r)=rs(x),g(x,r)=rs(x)+a(x,r),0<r<1,h(x,r)=r^{s(x)}, \qquad g(x,r)=r^{s(x)+a(x,r)}, \qquad0<r<1,

we obtain the sharper conclusion that

sup⁡x∈X0<r≤δ∣a(x,r)log⁡1r−λ∣⟶0⟹Hg=e−λHh.\sup_{\substack{x\in X\\0<r\leq\delta}}\left|a(x,r)\log\frac{1}{r}-\lambda\right|\longrightarrow0 \quad\Longrightarrow\quad\mathcal{H}^{g}=e^{-\lambda}\mathcal{H}^{h}.

Thus the scale 1/log⁡(1/r)1/\log(1/r) is the threshold for this uniform comparison principle.

  1. Failure of pointwise stability. We construct bounded gauges on [0,1][0,1] for which g(x,r)/h(x,r)→1g(x,r)/h(x,r)\to1 at every fixed point—in fact, the ratio is identically one for all sufficiently small rr, with the threshold depending on the point—while

Hg(E)=κHh(E)for every E⊆[0,1],\mathcal{H}^{g}(E)=\kappa\mathcal{H}^{h}(E) \qquad\text{for every } E\subseteq[0,1],

where κ∈(0,1)\kappa\in(0,1) is arbitrary. This shows that pointwise asymptotic equivalence alone cannot control the induced outer measure.

  1. Baire-category localization. If g(⋅,r)/h(⋅,r)g(\cdot,r)/h(\cdot,r) is continuous for every rr and converges pointwise to one on a Baire space, then for each η>0\eta>0 there is a dense open set on which the convergence is locally uniform to within η\eta. Intersecting these sets gives a dense GδG_{\delta} set GG. We then prove equality Hg(K)=Hh(K)\mathcal{H}^{g}(K)=\mathcal{H}^{h}(K) for every compact K⊆GK\subseteq G. More generally, a two-sided estimate holds on compact subsets of the corresponding η\eta-localization set.

The last result is deliberately a localization theorem, not a theorem for all Borel sets. The exceptional set X∖GX\setminus G is meagre, but it need not be negligible for the outer measure under consideration. Consequently, category-smallness alone does not settle the pointwise-to-measure problem on arbitrary Borel sets. We formulate this remaining issue as an open problem in § 7.

The paper is organized as follows. § 2 introduces the spatial gauges and covering contents. § 3 establishes the metric outer-measure property. § 4 develops comparison principles, and Section 5 contains the logarithmic criterion. § 6 discusses measurability and regularity. The counterexample and the Baire-category localization theorem are proved in § 7. § 8 places these results in the context of spatially dependent and centred Hausdorff measures.

Conventions. We work in [0,∞][0,\infty] with the usual order and the conventions a+∞=∞a+\infty=\infty, a⋅∞=∞a\cdot\infty=\infty for a>0a>0, 0⋅∞=00\cdot\infty=0, and inf⁡∅=∞\inf\varnothing=\infty. For A⊆XA\subseteq X we put diam⁡∅=0\operatorname{diam}\varnothing=0 and diam⁡A=sup⁡{d(x,y):x,y∈A}\operatorname{diam}A=\sup\{d(x,y):x,y\in A\} otherwise. For nonempty E,F⊆XE,F\subseteq X, dist⁡(E,F)=inf⁡{d(x,y):x∈E,y∈F}\operatorname{dist}(E,F)=\inf\{d(x,y):x\in E,y\in F\}. The Lebesgue outer measure on R\mathbb{R} is denoted by λ∗\lambda^*; we use its standard properties, including λ∗(I)=∣I∣\lambda^*(I)=|I| for every interval II, as developed for example in [7 Chapter 1].

Spatial gauges and covering constructions

Throughout, (X,d)(X,d) is a metric space.

Definition 2.1 (Spatial gauge). A spatial gauge is a function h:X×(0,∞)→[0,∞]h:X\times(0,\infty)\to[0,\infty]. No continuity, measurability or monotonicity is assumed unless explicitly stated.

Definition 2.2 (Covering cost and the measures Hδh\mathcal{H}^h_\delta, Hh\mathcal{H}^h). For A⊆XA\subseteq X define

ch(A)={inf⁡x∈Ah(x,diam⁡A),A≠∅, 0<diam⁡A<∞,0,A=∅ or diam⁡A=0.c_h(A)= \begin{cases} \inf_{x\in A} h(x,\operatorname{diam} A), & A\ne\varnothing,\ 0<\operatorname{diam} A<\infty,\\ 0, & A=\varnothing\text{ or }\operatorname{diam} A=0. \end{cases}

(Sets of infinite diameter never occur in what follows.) For δ>0\delta>0, a sequence (Aj)j≥1(A_j)_{j\geq1} of subsets of XX is a δ\delta-covering of E⊆XE\subseteq X if E⊆⋃jAjE\subseteq\bigcup_j A_j and diam⁡Aj≤δ\operatorname{diam} A_j\leq\delta for all jj. Set

Hδh(E)=inf⁡{∑j=1∞ch(Aj):(Aj)j≥1 is a δ-covering of E},Hh(E)=lim⁡δ↓0Hδh(E).\mathcal{H}^h_\delta(E)=\inf\left\{\sum_{j=1}^{\infty}c_h(A_j):(A_j)_{j\geq1}\text{ is a }\delta\text{-covering of }E\right\},\qquad\mathcal{H}^h(E)=\lim_{\delta\downarrow0}\mathcal{H}^h_\delta(E).

Finite coverings are included, since a finite family may be completed by empty sets, which have diameter and cost 00. If EE admits no countable δ\delta-covering (which can happen only if XX is not separable), then Hδh(E)=inf⁡∅=∞\mathcal{H}^h_\delta(E)=\inf\varnothing=\infty.

Lemma 2.3 (Existence of the limit). For every E⊆XE\subseteq X the function δ↦Hδh(E)\delta\mapsto\mathcal{H}^h_\delta(E) is nonincreasing on (0,∞)(0,\infty). Consequently the limit Hh(E)\mathcal{H}^h(E) exists in [0,∞][0,\infty] and Hh(E)=sup⁡δ>0Hδh(E)\mathcal{H}^h(E)=\sup_{\delta>0}\mathcal{H}^h_\delta(E).

Proof. Let 0<δ1<δ20<\delta_1<\delta_2. Every δ1\delta_1-covering of EE is a δ2\delta_2-covering of EE, so the infimum defining Hδ2h(E)\mathcal{H}^h_{\delta_2}(E) is taken over a larger class than the one defining Hδ1h(E)\mathcal{H}^h_{\delta_1}(E). Hence Hδ2h(E)≤Hδ1h(E)\mathcal{H}^h_{\delta_2}(E)\leq\mathcal{H}^h_{\delta_1}(E). A nonincreasing function of δ\delta with values in [0,∞][0,\infty] has a limit as δ↓0\delta\downarrow0, equal to its supremum.

Remark 2.4 (Sets of diameter zero). The convention ch(A)=0c_h(A)=0 when diam⁡A=0\operatorname{diam} A=0 is the usual Hausdorff-type convention for gauges satisfying h(x,r)→0h(x,r)\to0 as r↓0r\downarrow0; it is imposed here explicitly so that no value h(x,0)h(x,0) is needed. A set of diameter zero is either empty or a singleton. Consequently every countable set E={x1,x2,…}E=\{x_1,x_2,\ldots\} satisfies Hh(E)=0\mathcal{H}^h(E)=0: for every δ>0\delta>0 the sequence of singletons ({xj})j≥1(\{x_j\})_{j\geq1} is a δ\delta-covering of EE of total cost $0,so, so\mathcal{H}^h_\delta(E)=0.

The next two lemmas are used repeatedly.

Lemma 2.5 (Locality in the radius). Let h,gh,g be spatial gauges and r0>0r_0>0 with h(x,r)=g(x,r)h(x,r)=g(x,r) for all x∈Xx\in X and 0<r<r00<r<r_0. Then Hδh=Hδg\mathcal{H}^h_\delta=\mathcal{H}^g_\delta for 0<δ<r00<\delta<r_0, and hence Hh=Hg\mathcal{H}^h=\mathcal{H}^g. Moreover, Hκh=κHh\mathcal{H}^{\kappa h}=\kappa\mathcal{H}^h for every constant κ∈(0,∞)\kappa\in(0,\infty).

Proof. If 0<δ<r00 < \delta< r_0 and diam⁡A≤δ\operatorname{diam} A \le\delta, then either diam⁡A=0\operatorname{diam} A = 0 and ch(A)=0=cg(A)c_h(A) = 0 = c_g(A), or 0<diam⁡A<r00 < \operatorname{diam} A < r_0 and ch(A)=cg(A)c_h(A) = c_g(A) because the two infima involve the same values. Hence the two infima defining Hδh(E)\mathcal{H}^h_\delta(E) and Hδg(E)\mathcal{H}^g_\delta(E) coincide, and Theorem 2.3 gives Hh(E)=lim⁡δ↓0Hδh(E)=lim⁡δ↓0Hδg(E)=Hg(E)\mathcal{H}^h(E) = \lim_{\delta\downarrow0}\mathcal{H}^h_\delta(E) = \lim_{\delta\downarrow0}\mathcal{H}^g_\delta(E) = \mathcal{H}^g(E). For the last assertion, cκh(A)=κch(A)c_{\kappa h}(A) = \kappa c_h(A) for every AA (an infimum commutes with multiplication by κ>0\kappa> 0), hence ∑jcκh(Aj)=κ∑jch(Aj)\sum_j c_{\kappa h}(A_j) = \kappa\sum_j c_h(A_j) for every covering; taking infima and then limits gives Hδκh=κHδh\mathcal{H}^{\kappa h}_\delta= \kappa\mathcal{H}^h_\delta and Hκh=κHh\mathcal{H}^{\kappa h} = \kappa\mathcal{H}^h.

In particular, a gauge need only be specified for small radii; we shall use this freely.

Lemma 2.6 (Closed coverings suffice). For every A⊆XA \subseteq X one has diam⁡A‾=diam⁡A\operatorname{diam} \overline{A} = \operatorname{diam} A and ch(A‾)≤ch(A)c_h(\overline{A}) \le c_h(A). Consequently, for every E⊆XE \subseteq X and δ>0\delta> 0,

Hδh(E)=inf⁡{∑jch(Fj):(Fj)j≥1 is a δ-covering of E by closed sets}.\mathcal{H}^h_\delta(E) = \inf\left\{\sum_j c_h(F_j) : (F_j)_{j\ge1}\text{ is a $\delta$-covering of $E$ by closed sets}\right\}.

Proof. If x,y∈A‾x,y \in\overline{A}, choose xn,yn∈Ax_n,y_n \in A with xn→xx_n \to x, yn→yy_n \to y; then d(x,y)=lim⁡d(xn,yn)≤diam⁡Ad(x,y) = \lim d(x_n,y_n) \le\operatorname{diam} A. Thus diam⁡A‾≤diam⁡A\operatorname{diam} \overline{A} \le\operatorname{diam} A, and the reverse inequality is obvious. If diam⁡A=0\operatorname{diam} A = 0, then AA is empty or a singleton, hence closed, and A‾=A\overline{A} = A. If r=diam⁡A>0r = \operatorname{diam} A > 0, then diam⁡A‾=r\operatorname{diam} \overline{A} = r and, since A⊆A‾A \subseteq\overline{A},

ch(A‾)=inf⁡x∈A‾h(x,r)≤inf⁡x∈Ah(x,r)=ch(A).c_h(\overline{A}) = \inf_{x\in\overline{A}} h(x,r) \le\inf_{x\in A} h(x,r) = c_h(A).

Let II denote the infimum over closed δ\delta-coverings. Every closed δ\delta-covering is a δ\delta-covering, so Hδh(E)≤I\mathcal{H}^h_\delta(E) \le I. Conversely, if (Aj)(A_j) is any δ\delta-covering of EE, then (Aj‾)(\overline{A_j}) is a closed δ\delta-covering of EE with ∑jch(Aj‾)≤∑jch(Aj)\sum_j c_h(\overline{A_j}) \le\sum_j c_h(A_j), so I≤∑jch(Aj)I \le\sum_j c_h(A_j); taking the infimum over (Aj)(A_j) gives I≤Hδh(E)I \le\mathcal{H}^h_\delta(E).

Remark 2.7. The inequality ch(A‾)≤ch(A)c_h(\overline{A}) \le c_h(A) is a feature of the present construction, in which the location parameter is taken from the covering set. It fails for centred constructions, where the centres must lie in the set being measured [11, 12]; this is the source of the difficulty of Borel regularity for centred Hausdorff measures (see § 8).

The metric outer-measure theorem

Recall that an outer measure on XX is a map μ:P(X)→[0,∞]\mu:\mathcal{P}(X) \to[0,\infty] with μ(∅)=0\mu(\varnothing) = 0, which is monotone and countably subadditive; a set MM is μ\mu-measurable (in the sense of Carathéodory) if μ(T)=μ(T∩M)+μ(T∖M)\mu(T) = \mu(T \cap M) + \mu(T \setminus M) for every T⊆XT \subseteq X. An outer measure is metric if μ(E∪F)=μ(E)+μ(F)\mu(E \cup F) = \mu(E) + \mu(F) whenever dist⁡(E,F)>0\operatorname{dist}(E,F) > 0.

Theorem 3.1 (Metric Carathéodory construction). For every spatial gauge hh, each Hδh\mathcal{H}^h_\delta (δ>0\delta> 0) is an outer measure, and Hh\mathcal{H}^h is a metric outer measure on XX. In particular, every Borel subset of XX is Hh\mathcal{H}^h-measurable in the sense of Carathéodory, and the restriction of Hh\mathcal{H}^h to the Borel σ\sigma-algebra is a countably additive measure.

Proof. Step 1: Hδh(∅)=0\mathcal{H}^h_\delta(\varnothing) = 0 and Hh(∅)=0\mathcal{H}^h(\varnothing) = 0. The sequence Aj=∅A_j = \varnothing (j≥1j \ge1) is a δ\delta-covering of ∅\varnothing of cost 00; hence Hδh(∅)=0\mathcal{H}^h_\delta(\varnothing) = 0 for every δ\delta, and Hh(∅)=0\mathcal{H}^h(\varnothing) = 0.

Step 2: monotonicity. If E⊆FE \subseteq F, every δ\delta-covering of FF is a δ\delta-covering of EE, so Hδh(E)≤Hδh(F)\mathcal{H}^h_\delta(E) \le\mathcal{H}^h_\delta(F) for each δ>0\delta> 0. Letting δ↓0\delta\downarrow0 gives Hh(E)≤Hh(F)\mathcal{H}^h(E) \le\mathcal{H}^h(F).

Step 3: countable subadditivity of Hδh\mathcal{H}^h_\delta. Fix δ>0\delta> 0 and sets Ek⊆XE_k \subseteq X, k∈Nk \in\mathbb{N}. If ∑kHδh(Ek)=∞\sum_k \mathcal{H}^h_\delta(E_k) = \infty there is nothing to prove. Otherwise every Hδh(Ek)\mathcal{H}^h_\delta(E_k) is finite; given ε>0\varepsilon> 0, by definition of the infimum we can choose, for each kk, a δ\delta-covering (Ak,j)j≥1(A_{k,j})_{j\ge1} of EkE_k such that

∑j≥1ch(Ak,j)≤Hδh(Ek)+ε2−k.(1)\sum_{j\ge1} c_h(A_{k,j}) \le\mathcal{H}^h_\delta(E_k) + \varepsilon2^{-k}. \tag*{(1)}

Enumerate the countable family {Ak,j:k,j≥1}\{A_{k,j}: k,j \ge1\} as a single sequence (for instance via a bijection N→N×N\mathbb{N} \to\mathbb{N} \times\mathbb{N}). It covers ⋃kEk\bigcup_k E_k, each of its members has diameter at most δ\delta, and, because the terms are nonnegative, its total cost does not depend on the enumeration. Therefore

Hδh(⋃kEk)≤∑k∑jch(Ak,j)≤∑kHδh(Ek)+ε∑k2−k=∑kHδh(Ek)+ε.\mathcal{H}^{h}_{\delta}\left(\bigcup_k E_k\right) \le\sum_k \sum_j c_h(A_{k,j}) \le\sum_k \mathcal{H}^{h}_{\delta}(E_k) + \varepsilon\sum_k 2^{-k} = \sum_k \mathcal{H}^{h}_{\delta}(E_k) + \varepsilon.

Letting ε↓0\varepsilon\downarrow0 yields Hδh(⋃kEk)≤∑kHδh(Ek)\mathcal{H}^{h}_{\delta}(\bigcup_k E_k) \le\sum_k \mathcal{H}^{h}_{\delta}(E_k). Together with Steps 1 and 2, this shows that Hδh\mathcal{H}^{h}_{\delta} is an outer measure.

Step 4: countable subadditivity of Hh\mathcal{H}^{h}. By Theorem 2.3, Hδh(Ek)≤Hh(Ek)\mathcal{H}^{h}_{\delta}(E_k) \le\mathcal{H}^{h}(E_k) for every kk and every δ>0\delta> 0. Hence, by Step 3,

Hδh(⋃kEk)≤∑kHδh(Ek)≤∑kHh(Ek)(δ>0).\mathcal{H}^{h}_{\delta}\left(\bigcup_k E_k\right) \le\sum_k \mathcal{H}^{h}_{\delta}(E_k) \le\sum_k \mathcal{H}^{h}(E_k) \qquad(\delta> 0).

The right-hand side does not depend on δ\delta; letting δ↓0\delta\downarrow0 on the left gives Hh(⋃kEk)≤∑kHh(Ek)\mathcal{H}^{h}(\bigcup_k E_k) \le\sum_k \mathcal{H}^{h}(E_k). Thus Hh\mathcal{H}^{h} is an outer measure.

Step 5: the metric property. Let E,F⊆XE,F \subseteq X with a:=dist⁡(E,F)>0a := \operatorname{dist}(E,F) > 0 and fix 0<δ<a0 < \delta< a. If E∪FE \cup F has no δ\delta-covering, then Hδh(E∪F)=∞\mathcal{H}^{h}_{\delta}(E \cup F) = \infty and the inequality (1) below is trivial. Otherwise, let (Aj)(A_j) be a δ\delta-covering of E∪FE \cup F. No AjA_j meets both EE and FF: if x∈Aj∩Ex \in A_j \cap E and y∈Aj∩Fy \in A_j \cap F, then a≤d(x,y)≤diam⁡Aj≤δ<aa \le d(x,y) \le\operatorname{diam} A_j \le\delta< a, a contradiction. Put

IE={j:Aj∩E≠∅},IF={j:Aj∩F≠∅}.I_E = \{j : A_j \cap E \ne\varnothing\}, \qquad I_F = \{j : A_j \cap F \ne\varnothing\}.

These index sets are disjoint; the subfamily (Aj)j∈IE(A_j)_{j\in I_E} covers EE and the subfamily (Aj)j∈IF(A_j)_{j\in I_F} covers FF, and both are δ\delta-coverings (after completion by empty sets). Since all costs are nonnegative,

∑jch(Aj)≥∑j∈IEch(Aj)+∑j∈IFch(Aj)≥Hδh(E)+Hδh(F).\sum_j c_h(A_j) \ge\sum_{j\in I_E} c_h(A_j) + \sum_{j\in I_F} c_h(A_j) \ge\mathcal{H}^{h}_{\delta}(E) + \mathcal{H}^{h}_{\delta}(F).

Taking the infimum over all δ\delta-coverings of E∪FE \cup F yields

Hδh(E∪F)≥Hδh(E)+Hδh(F)(2)\mathcal{H}^{h}_{\delta}(E \cup F) \ge\mathcal{H}^{h}_{\delta}(E) + \mathcal{H}^{h}_{\delta}(F) \tag*{(2)}

(0<δ<a)(0 < \delta< a).

The reverse inequality holds by Step 3, so equality holds in (2) for every δ<a\delta< a. Letting δ↓0\delta\downarrow0 (both sides are nondecreasing as δ\delta decreases, so the limit of the sum is the sum of the limits in [0,∞][0,\infty]) we obtain Hh(E∪F)=Hh(E)+Hh(F)\mathcal{H}^{h}(E \cup F) = \mathcal{H}^{h}(E) + \mathcal{H}^{h}(F).

Step 6: Borel sets. By Steps 1–5, Hh\mathcal{H}^{h} is a metric outer measure. By Theorem 3.2 below, every closed set is Hh\mathcal{H}^{h}-measurable. By Carathéodory’s theorem [1], [7 Theorem 1.11], the Hh\mathcal{H}^{h}-measurable sets form a σ\sigma-algebra on which Hh\mathcal{H}^{h} is countably additive. This σ\sigma-algebra contains all closed sets, hence all open sets, hence the σ\sigma-algebra they generate, which is the Borel σ\sigma-algebra.

For completeness we include the proof of Carathéodory’s criterion in the form we need.

Lemma 3.2 (Carathéodory’s criterion). Let μ\mu be a metric outer measure on XX. Then every closed set C⊆XC \subseteq X is μ\mu-measurable.

Proof. Let T⊆XT \subseteq X. By subadditivity μ(T)≤μ(T∩C)+μ(T∖C)\mu(T) \le\mu(T \cap C) + \mu(T \setminus C), so it suffices to prove

μ(T∩C)+μ(T∖C)≤μ(T),(3)\mu(T \cap C) + \mu(T \setminus C) \le\mu(T), \tag*{(3)}

and we may assume μ(T)<∞\mu(T) < \infty. If C=∅C = \varnothing, (3) is trivial, so let C≠∅C \ne\varnothing and write ρ(x)=dist⁡({x},C)\rho(x) = \operatorname{dist}(\{x\}, C). For n∈Nn \in\mathbb{N} put

Tn={x∈T∖C:ρ(x)≥1/n},Rk={x∈T∖C:1k+1≤ρ(x)<1k}(k∈N).T_n = \{x \in T \setminus C : \rho(x) \ge1/n\}, \qquad R_k = \left\{x \in T \setminus C : \frac{1}{k+1} \le\rho(x) < \frac{1}{k}\right\} \quad(k \in\mathbb{N}).

(a) Since dist⁡(T∩C,Tn)≥1/n>0\operatorname{dist}(T \cap C, T_n) \ge1/n > 0 (if both are nonempty), the metric property and monotonicity give μ(T∩C)+μ(Tn)=μ((T∩C)∪Tn)≤μ(T)\mu(T \cap C) + \mu(T_n) = \mu((T \cap C) \cup T_n) \le\mu(T) for every nn.

(b) Since CC is closed, ρ(x)>0\rho(x) > 0 for every x∉Cx \notin C; hence T∖C=Tn∪⋃k≥nRkT \setminus C = T_n \cup\bigcup_{k \ge n} R_k for every nn.

(c) If k≥j+2k \ge j + 2, then dist⁡(Rj,Rk)>0\operatorname{dist}(R_j, R_k) > 0: for x∈Rjx \in R_j and y∈Rky \in R_k, the triangle inequality gives ρ(x)≤d(x,y)+ρ(y)\rho(x) \le d(x,y) + \rho(y), so d(x,y)≥ρ(x)−ρ(y)>1j+1−1k≥1j+1−1j+2>0d(x,y) \ge\rho(x) - \rho(y) > \frac{1}{j+1} - \frac{1}{k} \ge\frac{1}{j+1} - \frac{1}{j+2} > 0. Consequently, for every NN, the sets R2,R4,…,R2NR_2, R_4, \ldots, R_{2N} are pairwise positively separated, and so are R1,R3,…,R2N−1R_1, R_3, \ldots, R_{2N-1}. Since the distance from a finite union to a set is the minimum of the distances, an induction on NN using the metric property gives

∑i=1Nμ(R2i)=μ(⋃i=1NR2i)≤μ(T),∑i=1Nμ(R2i−1)=μ(⋃i=1NR2i−1)≤μ(T).\sum_{i=1}^{N} \mu(R_{2i}) = \mu\left(\bigcup_{i=1}^{N} R_{2i}\right) \le\mu(T), \qquad\sum_{i=1}^{N} \mu(R_{2i-1}) = \mu\left(\bigcup_{i=1}^{N} R_{2i-1}\right) \le\mu(T).

Hence ∑k≥1μ(Rk)≤2μ(T)<∞\sum_{k \ge1} \mu(R_k) \le2\mu(T) < \infty.

(d) By (b) and subadditivity, μ(T∖C)≤μ(Tn)+∑k≥nμ(Rk)\mu(T \setminus C) \le\mu(T_n) + \sum_{k \ge n} \mu(R_k) for every nn. The tail of a convergent series tends to 00, and μ(Tn)\mu(T_n) is nondecreasing in nn and bounded by μ(T∖C)\mu(T \setminus C); therefore μ(T∖C)=lim⁡nμ(Tn)\mu(T \setminus C) = \lim_{n} \mu(T_n). Letting n→∞n \to\infty in (a) gives (3).

Remark 3.3 (What Theorem 3.1 does not say). The theorem does not require hh to be measurable in xx, continuous, or monotone in rr. It does not establish local finiteness or any inner regularity. Borel regularity, which does not follow from the metric property alone, is proved separately in Theorem 6.2; outer regularity by open sets fails in general (Theorem 6.4).

Comparison of spatial gauges

We first isolate the elementary comparison mechanism.

Lemma 4.1 (Comparison at a fixed scale). Let h,gh,g be spatial gauges, δ>0\delta> 0 and 0<m≤M<∞0 < m \le M < \infty such that

mh(x,r)≤g(x,r)≤Mh(x,r)for all x∈X, 0<r≤δ.m h(x,r) \le g(x,r) \le M h(x,r) \qquad\text{for all } x \in X,\ 0 < r \le\delta.

Then mch(A)≤cg(A)≤Mch(A)m c_h(A) \le c_g(A) \le M c_h(A) whenever diam⁡A≤δ\operatorname{diam} A \le\delta, and mHδh(E)≤Hδg(E)≤MHδh(E)m \mathcal{H}^{h}_{\delta}(E) \le\mathcal{H}^{g}_{\delta}(E) \le M \mathcal{H}^{h}_{\delta}(E) for every E⊆XE \subseteq X.

Proof. If diam⁡A=0\operatorname{diam} A = 0 all three costs vanish. Let 0<r=diam⁡A≤δ0 < r = \operatorname{diam} A \le\delta. For every x∈Ax \in A, g(x,r)≥mh(x,r)≥mch(A)g(x,r) \ge m h(x,r) \ge m c_h(A); taking the infimum over x∈Ax \in A gives cg(A)≥mch(A)c_g(A) \ge m c_h(A). Also cg(A)≤g(x,r)≤Mh(x,r)c_g(A) \le g(x,r) \le M h(x,r) for every x∈Ax \in A, and taking the infimum over x∈Ax \in A gives cg(A)≤Minf⁡x∈Ah(x,r)=Mch(A)c_g(A) \le M \inf_{x \in A} h(x,r) = M c_h(A).

Now let E⊆XE \subseteq X. If EE has no δ\delta-covering, all three quantities are ∞\infty (recall m>0m > 0). Otherwise, for every δ\delta-covering (Aj)(A_j) of EE,

∑jcg(Aj)≥m∑jch(Aj)≥mHδh(E),Hδg(E)≤∑jcg(Aj)≤M∑jch(Aj).\sum_j c_g(A_j) \ge m \sum_j c_h(A_j) \ge m \mathcal{H}^{h}_{\delta}(E), \qquad\mathcal{H}^{g}_{\delta}(E) \le\sum_j c_g(A_j) \le M \sum_j c_h(A_j).

Taking the infimum over all δ\delta-coverings in the first chain gives Hδg(E)≥mHδh(E)\mathcal{H}^{g}_{\delta}(E) \ge m\mathcal{H}^{h}_{\delta}(E); in the second chain it gives Hδg(E)≤MHδh(E)\mathcal{H}^{g}_{\delta}(E) \le M\mathcal{H}^{h}_{\delta}(E), because inf⁡(Mt)=Minf⁡t\inf(Mt) = M\inf t for M∈(0,∞)M \in(0,\infty).

Lemma 4.2 (Passage to the limit). Let h,gh,g be spatial gauges, δ0>0\delta_0>0 and m,M:(0,δ0]→(0,∞)m,M:(0,\delta_0]\to(0,\infty) with m(δ)→1m(\delta)\to1 and M(δ)→1M(\delta)\to1 as δ↓0\delta\downarrow0, such that m(δ)Hδh(E)≤Hδg(E)≤M(δ)Hδh(E)m(\delta)\mathcal{H}^h_\delta(E)\le\mathcal{H}^g_\delta(E)\le M(\delta)\mathcal{H}^h_\delta(E) for all E⊆XE\subseteq X and 0<δ≤δ00<\delta\le\delta_0. Then Hg=Hh\mathcal{H}^g=\mathcal{H}^h.

Proof. Fix EE. If Hh(E)<∞\mathcal{H}^h(E)<\infty, then Hδh(E)→Hh(E)\mathcal{H}^h_\delta(E)\to\mathcal{H}^h(E) in [0,∞)[0,\infty), so both m(δ)Hδh(E)m(\delta)\mathcal{H}^h_\delta(E) and M(δ)Hδh(E)M(\delta)\mathcal{H}^h_\delta(E) converge to Hh(E)\mathcal{H}^h(E), and the squeeze gives Hg(E)=Hh(E)\mathcal{H}^g(E)=\mathcal{H}^h(E). If Hh(E)=∞\mathcal{H}^h(E)=\infty, then Hδh(E)→∞\mathcal{H}^h_\delta(E)\to\infty; since m(δ)≥12m(\delta)\ge\frac{1}{2} for small δ\delta, the lower bound forces Hδg(E)→∞\mathcal{H}^g_\delta(E)\to\infty, i.e. Hg(E)=∞\mathcal{H}^g(E)=\infty. (Symmetrically, Hg(E)=∞\mathcal{H}^g(E)=\infty forces Hh(E)=∞\mathcal{H}^h(E)=\infty through the upper bound, so the two cases are exhaustive.)

Theorem 4.3 (Uniform asymptotic equivalence). Let h,gh,g be spatial gauges. Suppose there are r0>0r_0>0 and a function ε:(0,r0)→[0,1)\varepsilon:(0,r_0)\to[0,1) with lim⁡r↓0ε(r)=0\lim_{r\downarrow0}\varepsilon(r)=0 such that

(1−ε(r))h(x,r)≤g(x,r)≤(1+ε(r))h(x,r)for all x∈X, 0<r<r0.(1-\varepsilon(r))h(x,r)\le g(x,r)\le(1+\varepsilon(r))h(x,r)\qquad\text{for all }x\in X,\ 0<r<r_0.

Then Hh(E)=Hg(E)\mathcal{H}^h(E)=\mathcal{H}^g(E) for every E⊆XE\subseteq X.

Proof. For 0<δ<r00<\delta<r_0 put η(δ)=sup⁡0<r≤δε(r)∈[0,1]\eta(\delta)=\sup_{0<r\le\delta}\varepsilon(r)\in[0,1]. The function η\eta is nondecreasing, and η(δ)→0\eta(\delta)\to0 as δ↓0\delta\downarrow0: given θ>0\theta>0 there is ρ>0\rho>0 with ε(r)<θ\varepsilon(r)<\theta for 0<r<ρ0<r<\rho, hence η(δ)≤θ\eta(\delta)\le\theta for δ<ρ\delta<\rho. Choose δ0<r0\delta_0<r_0 with η(δ0)≤12\eta(\delta_0)\le\frac{1}{2}. For 0<δ≤δ00<\delta\le\delta_0 and 0<r≤δ0<r\le\delta we have ε(r)≤η(δ)\varepsilon(r)\le\eta(\delta), hence

(1−η(δ))h(x,r)≤g(x,r)≤(1+η(δ))h(x,r)(x∈X).(1-\eta(\delta))h(x,r)\le g(x,r)\le(1+\eta(\delta))h(x,r)\qquad(x\in X).

Theorem 4.1 with m=1−η(δ)≥12m=1-\eta(\delta)\ge\frac{1}{2} and M=1+η(δ)M=1+\eta(\delta) gives (1−η(δ))Hδh≤Hδg≤(1+η(δ))Hδh(1-\eta(\delta))\mathcal{H}^h_\delta\le\mathcal{H}^g_\delta\le(1+\eta(\delta))\mathcal{H}^h_\delta, and Theorem 4.2 yields Hg=Hh\mathcal{H}^g=\mathcal{H}^h, including the case in which the common value is infinite.

Corollary 4.4 (Two-sided comparability). Suppose there are constants 0<m≤M<∞0<m\le M<\infty and r0>0r_0>0 such that mh(x,r)≤g(x,r)≤Mh(x,r)mh(x,r)\le g(x,r)\le Mh(x,r) for all x∈Xx\in X and 0<r<r00<r<r_0. Then

mHh(E)≤Hg(E)≤MHh(E)for every E⊆X.m\mathcal{H}^h(E)\le\mathcal{H}^g(E)\le M\mathcal{H}^h(E)\qquad\text{for every }E\subseteq X.

Proof. For 0<δ<r00<\delta<r_0, Theorem 4.1 gives mHδh(E)≤Hδg(E)≤MHδh(E)m\mathcal{H}^h_\delta(E)\le\mathcal{H}^g_\delta(E)\le M\mathcal{H}^h_\delta(E). Since Hδh(E)↑Hh(E)\mathcal{H}^h_\delta(E)\uparrow\mathcal{H}^h(E) and Hδg(E)↑Hg(E)\mathcal{H}^g_\delta(E)\uparrow\mathcal{H}^g(E) as δ↓0\delta\downarrow0, and multiplication by a positive constant is continuous and nondecreasing on [0,∞][0,\infty], letting δ↓0\delta\downarrow0 gives the claim.

Power gauges and the logarithmic threshold

We now specialize to power-type gauges. Let s:X→(0,∞)s:X\to(0,\infty) and, for 0<r<10<r<1, put h(x,r)=rs(x)h(x,r)=r^{s(x)}. Let a:X×(0,1)→Ra:X\times(0,1)\to\mathbb{R} and put g(x,r)=rs(x)+a(x,r)g(x,r)=r^{s(x)+a(x,r)} for 0<r<10<r<1. For r≥1r\ge1 both gauges may be defined arbitrarily: by Theorem 2.5 this does not affect Hh\mathcal{H}^h or Hg\mathcal{H}^g. (The condition s(x)+a(x,r)>0s(x)+a(x,r)>0, natural if one wants g(x,r)→0g(x,r)\to0, is not needed below.) The basic identity is

g(x,r)h(x,r)=ra(x,r)=exp⁡(a(x,r)log⁡r)=exp⁡(−a(x,r)log⁡1r),0<r<1.(4)\frac{g(x,r)}{h(x,r)}=r^{a(x,r)}=\exp(a(x,r)\log r)=\exp\left(-a(x,r)\log\frac{1}{r}\right),\qquad0<r<1. \tag*{(4)}

Theorem 5.1 (Logarithmic comparison). Let λ∈R\lambda\in\mathbb{R} and assume

θ(δ):=sup⁡{∣a(x,r)log⁡1r−λ∣:x∈X, 0<r≤δ}⟶0(δ↓0).\theta(\delta):=\sup\left\{\left|a(x,r)\log\frac{1}{r}-\lambda\right|:x\in X,\ 0<r\le\delta\right\}\longrightarrow0\qquad(\delta\downarrow0).

Then Hg(E)=e−λHh(E)\mathcal{H}^g(E)=e^{-\lambda}\mathcal{H}^h(E) for every $E\subseteq X.

Proof. Let h~=e−λh\tilde{h}=e^{-\lambda}h. By (4), g(x,r)/h~(x,r)=exp⁡(−(a(x,r)log⁡1r−λ))g(x,r)/\tilde{h}(x,r)=\exp\left(-\left(a(x,r)\log\frac{1}{r}-\lambda\right)\right) for 0<r<10<r<1. Choose δ0∈(0,1)\delta_0\in(0,1) with θ(δ0)<∞\theta(\delta_0)<\infty. For 0<δ≤δ00<\delta\le\delta_0, x∈Xx\in X and 0<r≤δ0<r\le\delta, the exponent has absolute value at most θ(δ)\theta(\delta), so

e−θ(δ)h~(x,r)≤g(x,r)≤eθ(δ)h~(x,r).e^{-\theta(\delta)}\tilde{h}(x,r)\le g(x,r)\le e^{\theta(\delta)}\tilde{h}(x,r).

(Here h~(x,r)∈(0,∞)\tilde{h}(x,r)\in(0,\infty).) Theorem 4.1 gives e−θ(δ)Hδh~≤Hδg≤eθ(δ)Hδh~e^{-\theta(\delta)}\mathcal{H}^{\tilde{h}}_\delta\le\mathcal{H}^{g}_\delta\le e^{\theta(\delta)}\mathcal{H}^{\tilde{h}}_\delta, and since e±θ(δ)→1e^{\pm\theta(\delta)}\to1, Theorem 4.2 yields Hg=Hh~\mathcal{H}^{g}=\mathcal{H}^{\tilde{h}}. Finally Hh~=e−λHh\mathcal{H}^{\tilde{h}}=e^{-\lambda}\mathcal{H}^{h} by Theorem 2.5.

Corollary 5.2 (Logarithmic stability criterion). If

lim⁡δ↓0sup⁡x∈X, 0<r≤δ∣a(x,r)∣log⁡1r=0,\lim_{\delta\downarrow0}\sup_{x\in X,\,0<r\le\delta}|a(x,r)|\log\frac{1}{r}=0,

then Hg(E)=Hh(E)\mathcal{H}^{g}(E)=\mathcal{H}^{h}(E) for every E⊆XE\subseteq X.

Proof. This is Theorem 5.1 with λ=0\lambda=0. Alternatively, put ω(δ)=sup⁡x, 0<r≤δ∣a(x,r)∣log⁡1r\omega(\delta)=\sup_{x,\,0<r\le\delta}|a(x,r)|\log\frac{1}{r}; then (4) gives e−ω(δ)≤g/h≤eω(δ)e^{-\omega(\delta)}\le g/h\le e^{\omega(\delta)} for 0<r≤δ0<r\le\delta, and one concludes with Theorems 4.1 and 4.2.

To evaluate the measures in the example below we need the classical identification of the one-dimensional measure on the line.

Lemma 5.3. Let X⊆RX\subseteq\mathbb{R} be an interval with the Euclidean metric and h(x,r)=rh(x,r)=r. Then Hδh(E)=λ∗(E)\mathcal{H}^{h}_{\delta}(E)=\lambda^*(E) for every E⊆XE\subseteq X and every δ>0\delta>0; in particular Hh(E)=λ∗(E)\mathcal{H}^{h}(E)=\lambda^*(E), and Hh([0,1])=1\mathcal{H}^{h}([0,1])=1 when X=[0,1]X=[0,1].

Proof. Here ch(A)=diam⁡Ac_h(A)=\operatorname{diam}A for every A⊆XA\subseteq X.

Lower bound. Let (Aj)(A_j) be a δ\delta-covering of EE. For Aj≠∅A_j\ne\varnothing let Ij=[inf⁡Aj,sup⁡Aj]I_j=[\inf A_j,\sup A_j], a closed interval with Aj⊆IjA_j\subseteq I_j and ∣Ij∣=diam⁡Aj|I_j|=\operatorname{diam}A_j; for Aj=∅A_j=\varnothing let Ij=∅I_j=\varnothing. Then E⊆⋃jIjE\subseteq\bigcup_j I_j and, by monotonicity and countable subadditivity of λ∗\lambda^*, λ∗(E)≤∑jλ∗(Ij)=∑jdiam⁡Aj=∑jch(Aj)\lambda^*(E)\le\sum_j\lambda^*(I_j)=\sum_j\operatorname{diam}A_j=\sum_j c_h(A_j). Taking the infimum gives λ∗(E)≤Hδh(E)\lambda^*(E)\le\mathcal{H}^{h}_{\delta}(E).

Upper bound. We may assume λ∗(E)<∞\lambda^*(E)<\infty. Let ε>0\varepsilon>0 and choose open intervals JkJ_k with E⊆⋃kJkE\subseteq\bigcup_kJ_k and ∑k∣Jk∣≤λ∗(E)+ε\sum_k|J_k|\le\lambda^*(E)+\varepsilon; we may discard empty JkJ_k, and since ∑k∣Jk∣<∞\sum_k|J_k|<\infty each JkJ_k is bounded. Divide the closure of JkJ_k into NkN_k consecutive closed intervals Pk,1,…,Pk,NkP_{k,1},\ldots,P_{k,N_k} of equal length ∣Jk∣/Nk≤δ|J_k|/N_k\le\delta. The sets Pk,i∩XP_{k,i}\cap X are subsets of XX with diam⁡(Pk,i∩X)≤∣Pk,i∣≤δ\operatorname{diam}(P_{k,i}\cap X)\le|P_{k,i}|\le\delta, they cover EE, and

∑k,ich(Pk,i∩X)≤∑k∑i=1Nk∣Pk,i∣=∑k∣Jk∣≤λ∗(E)+ε.\sum_{k,i}c_h(P_{k,i}\cap X)\le\sum_k\sum_{i=1}^{N_k}|P_{k,i}|=\sum_k|J_k|\le\lambda^*(E)+\varepsilon.

Hence Hδh(E)≤λ∗(E)+ε\mathcal{H}^{h}_{\delta}(E)\le\lambda^*(E)+\varepsilon for every ε>0\varepsilon>0. Combining both bounds gives Hδh(E)=λ∗(E)\mathcal{H}^{h}_{\delta}(E)=\lambda^*(E) for every δ\delta, and the remaining assertions follow from λ∗([0,1])=1\lambda^*([0,1])=1 [7 Chapter 1].

:::

Example 5.4 (A perturbation at the logarithmic scale). Let X=[0,1]X=[0,1] with the Euclidean metric, h(x,r)=rh(x,r)=r, and fix a∈Ra\in\mathbb{R}. Put ra=min⁡{1,ea}r_a=\min\{1,e^a\} and define, for 0<r<ra0<r<r_a,

g(x,r)=r1+a/log⁡(1/r).g(x,r)=r^{1+a/\log(1/r)}.

(For 0<r<ra0<r<r_a we have log⁡(1/r)>max⁡{0,−a}\log(1/r)>\max\{0,-a\}, so the exponent 1+a/log⁡(1/r)1+a/\log(1/r) is positive; for r≥rar\ge r_a the gauge gg may be defined arbitrarily, by Theorem 2.5.) Since ra/log⁡(1/r)=exp⁡(alog⁡(1/r)log⁡r)=exp⁡(−a)r^{a/\log(1/r)}=\exp\left(\frac{a}{\log(1/r)}\log r\right)=\exp(-a), we have

g(x,r)=e−ar(0<r<ra).g(x,r)=e^{-a}r\qquad(0<r<r_a).

By Theorems 2.5 and 5.3, Hg(E)=e−aHh(E)=e−aλ∗(E)\mathcal{H}^{g}(E)=e^{-a}\mathcal{H}^{h}(E)=e^{-a}\lambda^{*}(E) for every E⊆[0,1]E\subseteq[0,1]. In particular,

Hh([0,1])=1,Hg([0,1])=e−a.\mathcal{H}^{h}([0,1])=1,\qquad\mathcal{H}^{g}([0,1])=e^{-a}.

Here the exponent perturbation a(x,r)=a/log⁡(1/r)a(x,r)=a/\log(1/r) tends to zero, but a(x,r)log⁡(1/r)≡aa(x,r)\log(1/r)\equiv a; this agrees with Theorem 5.1 for λ=a\lambda=a. Thus a perturbation of the exponent that tends to zero need not preserve the measure when it is of order 1/log⁡(1/r)1/\log(1/r).

Remark 5.5 (Sharpness). Theorem 5.2 and Theorem 5.4 identify the sharp scale for the uniform comparison method: perturbations of order o(1/log⁡(1/r))o(1/\log(1/r)), uniformly in xx, preserve the measure, whereas a perturbation equal to λ/log⁡(1/r)\lambda/\log(1/r) with λ≠0\lambda\ne0 multiplies it by e−λ≠1e^{-\lambda}\ne1 (Theorem 5.1). This is a sharpness statement for uniform criteria; it is not a classification of all possible nonuniform perturbations (see § 7).

Regularity: measurability, Borel regularity and open sets

Definition 6.1. An outer measure μ\mu on XX is Borel regular if every Borel set is μ\mu-measurable and for every E⊆XE\subseteq X there exists a Borel set B⊇EB\supseteq E with μ(B)=μ(E)\mu(B)=\mu(E).

The metric property (Theorem 3.1) yields the first requirement but, by itself, gives no Borel hull: the proof of Theorem 3.1 constructs no approximation from above. The hull is provided by the closure lemma (Theorem 2.6).

Theorem 6.2 (Borel regularity). For every spatial gauge hh the outer measure Hh\mathcal{H}^{h} is Borel regular. More precisely, for every E⊆XE\subseteq X there is a set B⊇EB\supseteq E which is a countable intersection of countable unions of closed sets (an FσδF_{\sigma\delta} set) such that Hh(B)=Hh(E)\mathcal{H}^{h}(B)=\mathcal{H}^{h}(E).

Proof. By Theorem 3.1 every Borel set is Hh\mathcal{H}^{h}-measurable. Let E⊆XE\subseteq X. If Hh(E)=∞\mathcal{H}^{h}(E)=\infty, take B=XB=X (which is closed), since Hh(X)≥Hh(E)=∞\mathcal{H}^{h}(X)\ge\mathcal{H}^{h}(E)=\infty. Assume Hh(E)<∞\mathcal{H}^{h}(E)<\infty; then Hδh(E)≤Hh(E)<∞\mathcal{H}^{h}_{\delta}(E)\le\mathcal{H}^{h}(E)<\infty for every δ\delta (Theorem 2.3). For each n∈Nn\in\mathbb{N}, by Theorem 2.6 there is a 1/n1/n-covering (Fn,j)j≥1(F_{n,j})_{j\ge1} of EE by closed sets with

∑jch(Fn,j)≤H1/nh(E)+1n≤Hh(E)+1n.\sum_j c_h(F_{n,j})\le\mathcal{H}^{h}_{1/n}(E)+\frac{1}{n}\le\mathcal{H}^{h}(E)+\frac{1}{n}.

Put Bn=⋃jFn,jB_n=\bigcup_jF_{n,j} and B=⋂nBnB=\bigcap_nB_n. Then BB is FσδF_{\sigma\delta}, hence Borel, and E⊆BE\subseteq B because E⊆BnE\subseteq B_n for every nn. Since B⊆BnB\subseteq B_n, the family (Fn,j)j(F_{n,j})_j is a 1/n1/n-covering of BB, whence

H1/nh(B)≤∑jch(Fn,j)≤Hh(E)+1n(n∈N).\mathcal{H}^{h}_{1/n}(B)\le\sum_j c_h(F_{n,j})\le\mathcal{H}^{h}(E)+\frac{1}{n}\qquad(n\in\mathbb{N}).

Fix m∈Nm\in\mathbb{N}. For n≥mn\ge m, Theorem 2.3 gives H1/mh(B)≤H1/nh(B)≤Hh(E)+1n\mathcal{H}^{h}_{1/m}(B)\le\mathcal{H}^{h}_{1/n}(B)\le\mathcal{H}^{h}(E)+\frac{1}{n}; letting n→∞n\to\infty we get H1/mh(B)≤Hh(E)\mathcal{H}^{h}_{1/m}(B)\le\mathcal{H}^{h}(E), and letting m→∞m\to\infty we get Hh(B)≤Hh(E)\mathcal{H}^{h}(B)\le\mathcal{H}^{h}(E). The reverse inequality is monotonicity. □\square

As usual, Borel regularity yields continuity along increasing sequences of arbitrary sets.

Corollary 6.3. If E1⊆E2⊆⋯⊆XE_1\subseteq E_2\subseteq\cdots\subseteq X, then Hh(⋃nEn)=lim⁡n→∞Hh(En)\mathcal{H}^{h}\left(\bigcup_nE_n\right)=\lim_{n\to\infty}\mathcal{H}^{h}(E_n).

Proof. By Theorem 6.2 choose Borel sets Bk⊇EkB_k\supseteq E_k with Hh(Bk)=Hh(Ek)\mathcal{H}^{h}(B_k)=\mathcal{H}^{h}(E_k), and put Cn=⋂k≥nBkC_n=\bigcap_{k\ge n}B_k. Each CnC_n is Borel, Cn⊆Cn+1C_n\subseteq C_{n+1}, and En⊆Cn⊆BnE_n\subseteq C_n\subseteq B_n (because En⊆Ek⊆BkE_n\subseteq E_k\subseteq B_k for k≥nk\ge n); hence Hh(Cn)=Hh(En)\mathcal{H}^h(C_n)=\mathcal{H}^h(E_n). Since Hh\mathcal{H}^h is countably additive on Borel sets (Theorem 3.1), it is continuous along increasing sequences of Borel sets, and therefore

Hh(⋃nEn)≤Hh(⋃nCn)=lim⁡nHh(Cn)=lim⁡nHh(En)≤Hh(⋃nEn),\mathcal{H}^h\left(\bigcup_n E_n\right) \leq\mathcal{H}^h\left(\bigcup_n C_n\right) = \lim_n \mathcal{H}^h(C_n) = \lim_n \mathcal{H}^h(E_n) \leq\mathcal{H}^h\left(\bigcup_n E_n\right),

where the limits exist by monotonicity.

Borel regularity must not be confused with approximation from above by open sets.

Example 6.4 (No outer regularity by open sets). Let X=RX=\mathbb{R} and h(x,r)=r1/2h(x,r)=r^{1/2}. Then Hh({0})=0\mathcal{H}^h(\{0\})=0 by Theorem 2.4, but Hh(U)=∞\mathcal{H}^h(U)=\infty for every open set U∋0U\ni0. Indeed, UU contains a nondegenerate compact interval II. Let 0<δ<10<\delta<1 and let (Aj)(A_j) be a δ\delta-covering of II, with dj=diam⁡Aj≤δd_j=\operatorname{diam} A_j\leq\delta. Since dj1/2≥δ−1/2djd_j^{1/2}\geq\delta^{-1/2}d_j for 0≤dj≤δ0\leq d_j\leq\delta, and by the lower bound in the proof of Theorem 5.3,

∑jch(Aj)=∑jdj1/2≥δ−1/2∑jdj≥δ−1/2λ∗(I)=δ−1/2∣I∣.\sum_j c_h(A_j)=\sum_j d_j^{1/2}\geq\delta^{-1/2}\sum_j d_j\geq\delta^{-1/2}\lambda^*(I)=\delta^{-1/2}|I|.

Hence Hδh(I)≥δ−1/2∣I∣→∞\mathcal{H}^h_\delta(I)\geq\delta^{-1/2}|I|\to\infty as δ↓0\delta\downarrow0, and Hh(U)≥Hh(I)=∞\mathcal{H}^h(U)\geq\mathcal{H}^h(I)=\infty.

Remark 6.5. For spatial gauges, approximation by open sets is delicate for a structural reason as well: replacing a covering set AA by an open neighbourhood Aρ={x:dist⁡({x},A)<ρ}A_\rho=\{x:\operatorname{dist}(\{x\},A)<\rho\} increases the diameter (up to diam⁡A+2ρ\operatorname{diam} A+2\rho), and since hh need not be monotone or continuous in rr, the cost ch(Aρ)c_h(A_\rho) is not controlled by ch(A)c_h(A). Continuity in the spatial variable and suitable uniform control in the radius are natural hypotheses for results of this type; we do not pursue them here.

The local stability problem

The uniform comparison in Theorem 4.3 is strong. It is natural to ask whether it can be weakened to pointwise asymptotic equivalence.

Open problem 7.1 (Pointwise-to-measure stability). Suppose h,g:X×(0,r0)→(0,∞)h,g:X\times(0,r_0)\to(0,\infty) satisfy

lim⁡r↓0g(x,r)h(x,r)=1\lim_{r\downarrow0}\frac{g(x,r)}{h(x,r)}=1

for every fixed x∈Xx\in X.

Find additional hypotheses, as weak as possible, under which Hg(E)=Hh(E)\mathcal{H}^g(E)=\mathcal{H}^h(E) for every Borel set E⊆XE\subseteq X.

The difficulty is that the covering cost is an infimum over all points of a covering set. Pointwise convergence at each fixed point does not provide a comparison uniform over all points that may occur in a small covering set. The following example shows that additional hypotheses are indeed indispensable, even on a compact space and for bounded gauges.

Proposition 7.2 (Pointwise equivalence does not suffice). Let X=[0,1]X=[0,1] with the Euclidean metric and fix κ∈(0,1)\kappa\in(0,1). For 0<r<10<r<1 let DrD_r be the set of rational numbers p/q∈[0,1]p/q\in[0,1], written in lowest terms with q≥1q\geq1, such that q≥1/rq\geq1/r. Define

h(x,r)=r,g(x,r)={κr,x∈Dr,r,x∉Dr,(x∈[0,1], 0<r<1).h(x,r)=r,\qquad g(x,r)= \begin{cases} \kappa r, & x\in D_r,\\ r, & x\notin D_r, \end{cases} \qquad(x\in[0,1],\ 0<r<1).

Then 0<g≤h≤10 < g \le h \le1, and for every x∈[0,1]x \in[0,1] there is ρ(x)>0\rho(x) > 0 with g(x,r)=h(x,r)g(x,r) = h(x,r) for all 0<r<ρ(x)0 < r < \rho(x); in particular g(x,r)/h(x,r)→1g(x,r)/h(x,r) \to1 for every xx. Nevertheless

Hg(E)=κλ∗(E)=κHh(E)for every E⊆[0,1];\mathcal{H}^{g}(E) = \kappa\lambda^*(E) = \kappa\mathcal{H}^{h}(E) \qquad\text{for every } E \subseteq[0,1];

in particular Hg([0,1])=κ<1=Hh([0,1])\mathcal{H}^{g}([0,1]) = \kappa< 1 = \mathcal{H}^{h}([0,1]).

Proof. Pointwise behaviour. If xx is irrational, then x∉Drx \notin D_r for all rr, and we take ρ(x)=1\rho(x) = 1. If x=p/qx = p/q in lowest terms, then x∈Drx \in D_r if and only if r≥1/qr \ge1/q; hence g(x,r)=h(x,r)g(x,r) = h(x,r) for 0<r<ρ(x):=1/q0 < r < \rho(x) := 1/q.

A density property. Let P⊆[0,1]P \subseteq[0,1] be an interval of length r∈(0,1)r \in(0,1). Then PP contains a point of DrD_r. Indeed, PP contains infinitely many rational numbers, whereas [0,1][0,1] contains only finitely many rationals whose reduced denominator is <1/r< 1/r (at most q+1q+1 for each of the finitely many q<1/rq < 1/r). Consequently cg(P)=inf⁡x∈Pg(x,r)=κr=κch(P)c_g(P) = \inf_{x \in P} g(x,r) = \kappa r = \kappa c_h(P), because g≥κhg \ge\kappa h everywhere and the value κr\kappa r is attained on P∩Dr≠∅P \cap D_r \ne\varnothing.

Lower bound. Since κh≤g≤h\kappa h \le g \le h on X×(0,1)X \times(0,1), Theorem 4.4 gives κHh(E)≤Hg(E)≤Hh(E)\kappa\mathcal{H}^{h}(E) \le\mathcal{H}^{g}(E) \le\mathcal{H}^{h}(E), and Hh=λ∗\mathcal{H}^{h} = \lambda^* on [0,1][0,1] by Theorem 5.3.

Upper bound. Let E⊆[0,1]E \subseteq[0,1], 0<δ<10 < \delta< 1 and ε>0\varepsilon> 0. As in the proof of Theorem 5.3, there are closed intervals Pk,iP_{k,i} of length at most δ\delta such that the sets Qk,i=Pk,i∩[0,1]Q_{k,i} = P_{k,i} \cap[0,1] cover EE and ∑k,i∣Pk,i∣≤λ∗(E)+ε\sum_{k,i} |P_{k,i}| \le\lambda^*(E) + \varepsilon. Each Qk,iQ_{k,i} is an interval contained in [0,1][0,1]; if it is degenerate, its cost is 00, and otherwise its length r=diam⁡Qk,i∈(0,δ]r = \operatorname{diam} Q_{k,i} \in(0,\delta] and, by the density property, cg(Qk,i)=κdiam⁡Qk,ic_g(Q_{k,i}) = \kappa\operatorname{diam} Q_{k,i}. Hence

Hδg(E)≤∑k,icg(Qk,i)≤κ∑k,i∣Pk,i∣≤κ(λ∗(E)+ε).\mathcal{H}^{g}_{\delta}(E) \le\sum_{k,i} c_g(Q_{k,i}) \le\kappa\sum_{k,i} |P_{k,i}| \le\kappa\bigl(\lambda^*(E) + \varepsilon\bigr).

Letting ε↓0\varepsilon\downarrow0 and then δ↓0\delta\downarrow0 gives Hg(E)≤κλ∗(E)\mathcal{H}^{g}(E) \le\kappa\lambda^*(E), which together with the lower bound proves the claim. □\square

In this example the ratio g(⋅,r)/h(⋅,r)g(\cdot,r)/h(\cdot,r) is discontinuous: for every rr it equals κ\kappa on the dense set DrD_r. For ratios that are continuous in the spatial variable, a Baire-category argument converts pointwise convergence into locally uniform convergence on a large set.

Proposition 7.3 (Baire localization). Let XX be a Baire space (for instance a complete metric space), and let h,g:X×(0,r0)→(0,∞)h,g : X \times(0,r_0) \to(0,\infty) be such that, for each r∈(0,r0)r \in(0,r_0), the ratio φr=g(⋅,r)/h(⋅,r)\varphi_r = g(\cdot,r)/h(\cdot,r) is continuous on XX, and φr(x)→1\varphi_r(x) \to1 as r↓0r \downarrow0 for every x∈Xx \in X. For η>0\eta> 0 let GηG_\eta be the set of points x∈Xx \in X having an open neighbourhood UU and a number ρ∈(0,r0)\rho\in(0,r_0) such that ∣φr−1∣≤η|\varphi_r - 1| \le\eta on UU for all 0<r<ρ0 < r < \rho. Then GηG_\eta is open and dense, and G=⋂k≥1G1/kG = \bigcap_{k \ge1} G_{1/k} is a dense GδG_\delta subset of XX.

Proof. GηG_\eta is open, because the neighbourhood UU in its definition is contained in GηG_\eta. To prove density, let V≠∅V \ne\varnothing be open. For n∈Nn \in\mathbb{N} with 1/n<r01/n < r_0 put

Fn={x∈V:∣φr(x)−1∣≤η for all 0<r<1/n}=V∩⋂0<r<1/nφr−1([1−η,1+η]).F_n = \{x \in V : |\varphi_r(x) - 1| \le\eta\text{ for all } 0 < r < 1/n\} = V \cap\bigcap_{0<r<1/n} \varphi_r^{-1}([1-\eta,1+\eta]).

By continuity of each φr\varphi_r, every FnF_n is relatively closed in VV; by pointwise convergence, V=⋃nFnV = \bigcup_n F_n. A nonempty open subspace of a Baire space is a Baire space, hence is not a countable union of nowhere dense subsets; therefore some FnF_n has nonempty interior WW relative to VV, and WW is open in XX because VV is. Every point of WW belongs to GηG_\eta (take U=WU = W and ρ=1/n\rho= 1/n), so Gη∩V≠∅G_\eta\cap V \ne\varnothing. Finally, GG is a countable intersection of dense open sets, hence a dense GδG_\delta set by the Baire property. □\square

Theorem 7.4 (Local stability on compact sets). Under the hypotheses of Theorem 7.3, Hg(K)=Hh(K)\mathcal{H}^{g}(K) = \mathcal{H}^{h}(K) for every compact set K⊆GK \subseteq G, and more generally (1−η)Hh(K)≤Hg(K)≤(1+η)Hh(K)(1-\eta)\mathcal{H}^{h}(K) \leq\mathcal{H}^{g}(K) \leq(1+\eta)\mathcal{H}^{h}(K) for every compact K⊆GηK \subseteq G_{\eta} with 0<η<10 < \eta< 1. Consequently Hg(E)=Hh(E)\mathcal{H}^{g}(E) = \mathcal{H}^{h}(E) for every set EE that is a countable increasing union of compact subsets of GG.

Proof. Let 0<η<10 < \eta< 1 and let K⊆GηK \subseteq G_{\eta} be compact and nonempty. Each x∈Kx \in K has an open neighbourhood UxU_x and ρx>0\rho_x > 0 as in the definition of GηG_{\eta}; by compactness, K⊆U:=Ux1∪⋯∪UxNK \subseteq U := U_{x_1} \cup\cdots\cup U_{x_N}, and with ρ=min⁡iρxi\rho= \min_i \rho_{x_i} we have ∣φr−1∣≤η|\varphi_r - 1| \leq\eta on UU for 0<r<ρ0 < r < \rho. Let τ=dist⁡(K,X∖U)\tau= \operatorname{dist}(K, X \setminus U) if U≠XU \ne X and τ=∞\tau= \infty otherwise; τ>0\tau> 0 because KK is compact, X∖UX \setminus U is closed and disjoint from KK, and the continuous function dist⁡(⋅,X∖U)\operatorname{dist}(\mathord{\cdot}, X \setminus U) attains a positive minimum on KK.

Fix 0<δ<min⁡{ρ,τ}0 < \delta< \min\{\rho,\tau\} and let (Aj)(A_j) be a δ\delta-covering of KK. Discard the sets that do not meet KK; the remaining family (Aj)j∈J(A_j)_{j \in J} still covers KK and has smaller total cost. Each AjA_j, j∈Jj \in J, is contained in UU: if x∈Aj∩Kx \in A_j \cap K and y∈Ajy \in A_j, then d(x,y)≤δ<τd(x,y) \leq\delta< \tau, so y∈Uy \in U. Hence for j∈Jj \in J with r=diam⁡Aj>0r = \operatorname{diam} A_j > 0 we have (1−η)h(y,r)≤g(y,r)≤(1+η)h(y,r)(1-\eta)h(y,r) \leq g(y,r) \leq(1+\eta)h(y,r) for all y∈Ajy \in A_j, and taking infima over y∈Ajy \in A_j as in the proof of Theorem 4.1, (1−η)ch(Aj)≤cg(Aj)≤(1+η)ch(Aj)(1-\eta)c_h(A_j) \leq c_g(A_j) \leq(1+\eta)c_h(A_j) for all j∈Jj \in J. Therefore

∑jcg(Aj)≥∑j∈Jcg(Aj)≥(1−η)∑j∈Jch(Aj)≥(1−η)Hδh(K).\sum_j c_g(A_j) \geq\sum_{j \in J} c_g(A_j) \geq(1-\eta)\sum_{j \in J} c_h(A_j) \geq(1-\eta)\mathcal{H}^{h}_{\delta}(K).
Hδg(K)≤∑j∈Jcg(Aj)≤(1+η)∑j∈Jch(Aj)≤(1+η)∑jch(Aj).\mathcal{H}^{g}_{\delta}(K) \leq\sum_{j \in J} c_g(A_j) \leq(1+\eta)\sum_{j \in J} c_h(A_j) \leq(1+\eta)\sum_j c_h(A_j).

Taking infima over all δ\delta-coverings of KK gives (1−η)Hδh(K)≤Hδg(K)≤(1+η)Hδh(K)(1-\eta)\mathcal{H}^{h}_{\delta}(K) \leq\mathcal{H}^{g}_{\delta}(K) \leq(1+\eta)\mathcal{H}^{h}_{\delta}(K) for all small δ\delta, and letting δ↓0\delta\downarrow0 yields the two-sided estimate. If K⊆GK \subseteq G, the estimate holds for every η=1/k\eta= 1/k, and letting k→∞k \to\infty gives Hg(K)=Hh(K)\mathcal{H}^{g}(K) = \mathcal{H}^{h}(K) (if one side is infinite, so is the other). The last assertion follows from Theorem 6.3, applied to Hg\mathcal{H}^{g} and to Hh\mathcal{H}^{h}.

Theorem 7.4 pinpoints where the difficulty of Theorem 7.1 lies for continuous ratios: in the closed nowhere dense set X∖GηX \setminus G_{\eta}. Such a set may nevertheless carry positive Hh\mathcal{H}^{h}-measure (think of a Cantor set of positive Lebesgue measure in [0,1][0,1]), so a positive answer requires controlling coverings near X∖GηX \setminus G_{\eta}.

Open problem 7.5 (Optimality and counterexamples). By Theorem 7.2, pointwise asymptotic equivalence does not imply equality of the induced outer measures, even when XX is compact and the gauges are uniformly bounded. Decide whether the implication holds when, in addition, the ratio g(⋅,r)/h(⋅,r)g(\mathord{\cdot},r)/h(\mathord{\cdot},r) is continuous on XX for each rr (or gg and hh are jointly continuous), and, if it fails, which further hypotheses restore it.

A successful solution should contain both a positive theorem and a counterexample showing that its hypotheses cannot be substantially weakened. Several approaches deserve investigation.

  1. Compact-uniform convergence. Assume that the ratio converges uniformly in xx on every compact subset. The proof of Theorem 7.4 shows that the points of a covering set lying outside a neighbourhood of the compact set never need to be considered at small scales; it then remains to pass from compact sets to Borel sets, for instance under inner regularity of Hh\mathcal{H}^{h} by compact sets.

(ii) Equicontinuity in the spatial variable. A common modulus of continuity for the ratios φr\varphi_r converts pointwise convergence into uniform convergence on compact sets, by a finite-net argument, and reduces the problem to (i). Without equicontinuity, Theorem 7.3 still yields locally uniform convergence on the dense open sets GηG_\eta.

(iii) Restricted covering families. If the construction is modified so that the location parameter is selected from the portion of the covering set lying in a prescribed compact region, localization may become possible. This defines a different construction and must be studied on its own terms.

(iv) Capacity or tightness assumptions. A condition preventing low-cost points from being used by coverings of geometrically distant regions may yield a useful substitute for uniform convergence; note that in Theorem 7.2 the low-cost sets DrD_r are dense at every scale.

Relation to previous work and scope of the results

The construction of metric outer measures by coverings is classical. Carathéodory’s method, the metric outer-measure principle, and the criterion ensuring measurability of closed sets are treated in standard references such as [6 §§ 2.3, 2.10], [10], [8 Chapters 1 and 4], [7], [14 Chapter 5], and [5]. Accordingly, the metric property proved in Theorem 3.1 is included to make the framework self-contained; it is not asserted as a new general principle. The same qualification applies to the elementary comparison argument in Theorem 4.3. The closure inequality in Theorem 2.6, which follows from taking the location parameter over the covering set itself, is the simple structural fact that makes the Borel-hull argument particularly direct in this setting.

Spatially dependent Hausdorff gauges have appeared in several forms. Gauges of the type h(x,r)=rtν(B(x,r))qh(x,r)=r^t\nu(B(x,r))^q, where ν\nu is a measure, occur in Olsen’s multifractal formalism [9]. Centred Hausdorff measures, introduced by Saint Raymond and Tricot [11], were studied further by Schechter [12, 13]. In centred constructions the admissible centre of a covering ball is constrained to belong to the set being measured. That constraint makes replacement of a covering set by its closure more delicate. In the construction used here, the location variable ranges over the covering set, and hence closure does not increase the cost. This distinction explains the unconditional Borel-regularity statement in Theorem 6.2; it should not be confused with a corresponding assertion for centred Hausdorff measures.

The stability questions considered here are different from the basic construction problem. The logarithmic criterion in Theorem 5.1 is a quantitative consequence of uniform comparison, and its proof is elementary. Its role is to identify the exact scale at which uniform exponent perturbations produce a nontrivial multiplicative change. The more substantial issue is the failure of pointwise stability: Theorem 7.2 shows that even eventual pointwise equality of the gauges at each fixed spatial point does not prevent a fixed change in the induced measure. The mechanism is that every interval at a given scale contains points where the covering cost is reduced, although the low-cost points depend on the scale.

The continuity assumption in Theorem 7.3 prevents that particular dense-set mechanism from operating without restriction. Baire’s theorem gives local uniform control on a dense open set at each fixed tolerance, and compactness then turns this control into comparison of covering costs. The resulting statement, Theorem 7.4, applies to compact subsets of a dense GδG_\delta set, not to all Borel sets. This limitation is essential to the formulation: a meagre exceptional set can still have positive measure, and the present argument does not control coverings that approach that exceptional set.

Thus the contribution of the paper is best understood as a set of precise stability statements and a counterexample delimiting their scope, rather than as a new construction of metric outer measure. The following question remains open.

Conclusions

We have studied the stability of Carathéodory outer measures generated by covering costs whose values depend on both scale and spatial position. The underlying construction yields a metric outer measure and is Borel regular; these foundational facts are included to clarify the setting in which the stability questions are posed. In particular, every set has an FσδF_{\sigma\delta} Borel hull of the same outer measure, whereas approximation from above by open sets may fail.

For uniformly controlled perturbations, the induced outer measures satisfy sharp comparison principles. In the power-gauge setting, the condition

sup⁡x∈X0<r≤δ∣a(x,r)log⁡1r−λ∣⟶0\sup_{\substack{x\in X\\0<r\le\delta}}\left|a(x,r)\log\frac{1}{r}-\lambda\right|\longrightarrow0

implies Hg=e−λHh\mathcal{H}^{g}=e^{-\lambda}\mathcal{H}^{h}. The logarithmic scale therefore identifies the threshold for this uniform criterion.

The principal nonuniform phenomenon is that pointwise asymptotic equivalence does not imply equality of the induced outer measures. The example on [0,1][0,1] gives bounded gauges whose ratio is eventually one at each fixed point, while the resulting measures differ by a fixed factor. Under spatial continuity of the ratio, Baire-category arguments recover equality on compact subsets of a dense GδG_\delta set. They do not, however, settle the question for all Borel sets, because the meagre exceptional set may carry positive measure.

A natural next step is to identify additional hypotheses—such as equicontinuity, compact-uniform convergence, or suitable control of the exceptional set—that yield stability on larger classes of sets. Any such result should be accompanied by counterexamples that clarify the sharpness of its assumptions.

References

  1. [1]C. Carathéodory, Über das lineare Maß von Punktmengen – eine Verallgemeinerung des Längenbegriffs, Nachrichten von der Gesellschaft der Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse (1914), 404–426.
  2. [2]C. Carathéodory, Vorlesungen über reelle Funktionen, 2nd ed., B. G. Teubner, Leipzig–Berlin, 1927 (1st ed. 1918).
  3. [5]Encyclopedia of Mathematics, Outer measure (section “Caratheodory constructions of metric outer measures”), EMS Press. https://encyclopediaofmath.org/wiki/Outer_measure.
  4. [6]H. Federer, Geometric Measure Theory, Grundlehren der mathematischen Wissenschaften 153, Springer-Verlag, New York, 1969.
  5. [7]G. B. Folland, Real Analysis: Modern Techniques and Their Applications, 2nd ed., John Wiley & Sons, New York, 1999.
  6. [8]P. Mattila, Geometry of Sets and Measures in Euclidean Spaces. Fractals and Rectifiability, Cambridge Studies in Advanced Mathematics 44, Cambridge University Press, Cambridge, 1995.
  7. [9]L. Olsen, A multifractal formalism, Advances in Mathematics 116 (1995), 82–196.DOI
  8. [10]C. A. Rogers, Hausdorff Measures, Cambridge University Press, Cambridge, 1970.
  9. [11]X. Saint Raymond and C. Tricot, Packing regularity of sets in n-space, Mathematical Proceedings of the Cambridge Philosophical Society 103 (1988), 133–145.DOI
  10. [12]A. Schechter, On the centred Hausdorff measure, Journal of the London Mathematical Society (2) 62 (2000), no. 3, 843–851. doi:10.1112/S0024610700001356.DOI
  11. [13]A. Schechter, Regularity and Other Properties of Hausdorff Measures, Ph.D. thesis, University College London, 2002. https://discovery.ucl.ac.uk/id/eprint/10102053/.
  12. [14]M. E. Taylor, Measure Theory and Integration, Graduate Studies in Mathematics 76, American Mathematical Society, Providence, RI, 2006.DOI

Paper details

vlnc:W_45bztv5q
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

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