Packing dimension of mean porous measures
D. Beliaev, E. J\"arvenp\"a\"a, M. J\"arvenp\"a\"a, A. K\"aenm\"aki,, T. Rajala, S. Smirnov, V. Suomala

TL;DR
This paper establishes an upper bound on the packing dimension of mean porous measures in Euclidean space, showing it depends on mean porosity and approaches the space dimension minus one as porosity increases.
Contribution
It provides a rigorous proof for the packing dimension bound of mean porous measures, correcting previous incorrect proofs and demonstrating that such measures are not necessarily approximable by mean porous sets.
Findings
Upper bound on packing dimension depends on mean porosity
Bound approaches d-1 as mean porosity reaches maximum
Existence of mean porous measure not approximable by mean porous sets
Abstract
We prove that the packing dimension of any mean porous Radon measure on may be estimated from above by a function which depends on mean porosity. The upper bound tends to as mean porosity tends to its maximum value. This result was stated in \cite{BS}, and in a weaker form in \cite{JJ1}, but the proofs are not correct. Quite surprisingly, it turns out that mean porous measures are not necessarily approximable by mean porous sets. We verify this by constructing an example of a mean porous measure on such that for all mean porous sets .
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Packing dimension of mean porous measures
D. Beliaev, E. Järvenpää, M. Järvenpää, A. Käenmäki, T. Rajala,
S. Smirnov and V. Suomala
DB: Department of Mathematics
Princeton University
Fine Hall, Washington Road Princeton NJ 08544-1000
USA
EJ,MJ,AK,TR and VS: Department of Mathematics and Statistics
P.O. Box 35 (MaD)
FIN-40014 University of Jyväskylä
Finland
SS: Department of Mathematics
University of Geneva
2-4 rue du Lièvre, Case postale 64
1211 Genève 4
Switzerland
Abstract.
We prove that the packing dimension of any mean porous Radon measure on may be estimated from above by a function which depends on mean porosity. The upper bound tends to as mean porosity tends to its maximum value. This result was stated in [BS], and in a weaker form in [JJ1], but the proofs are not correct. Quite surprisingly, it turns out that mean porous measures are not necessarily approximable by mean porous sets. We verify this by constructing an example of a mean porous measure on such that for all mean porous sets .
Key words and phrases:
Porosity, mean porosity, packing dimension, Radon measures
2000 Mathematics Subject Classification:
28A75, 28A80
EJ, MJ, AK, TR and VS acknowledge the support of the Academy of Finland (projects # 211229 and # 114821) and the Centre of Excellence in Analysis and Dynamics Research. EJ also thanks the hospitality of the University of Geneva. DB, EJ and SS acknowledge the support of the Swiss National Science Foundation. TR appreciates the financial support of Vilho, Yrjö and Kalle Väisälä foundation. Finally, we all thank the referee for valuable comments clarifying the exposition.
1. Introduction
Intuitively, it seems obvious that if a set contains relatively large holes at all small scales then the dimension of the set should be smaller than that of the ambient space. This observation was generalized and made into a quantitative form by Mattila [Ma1] in terms of a concept called porosity which describes the sizes of holes at all small scales (for the definition see Section 2). Mattila proved that if the porosity of a subset of is close to its maximum value then its Hausdorff dimension cannot be much bigger than . The correct asymptotic behaviour was established by Salli [S]. He also showed that Hausdorff dimension may be replaced by packing dimension, and moreover, by box counting dimension under the assumption that the set is uniformly porous.
The above mentioned result for Hausdorff dimension fails if the set contains large holes only at sequences of arbitrarily small scales; there are examples of such sets in with Hausdorff dimension [Ma2]. Nevertheless, the assumption that the set has relatively large holes at all small scales may be weakened to obtain an upper bound for Hausdorff dimension, or more generally, for packing dimension. In fact, it is sufficient to suppose that a certain percentage of scales contains holes. This leads to the concept of mean porosity (for the definition see Section 2). Dimensional properties of such sets were considered by Koskela and Rohde [KR] in the case of small mean porosity, and by Beliaev and Smirnov in the case of large one. For other related results, see [JJKS].
In this paper, the emphasis is given to packing dimensions of mean porous measures (for the definition see Section 2). Porous measures were introduced by Eckmann, Järvenpää and Järvenpää [EJJ] whilst the analogue of Mattila’s result was verified for porous measures in [JJ1]. Note that in [JJ1] the results are claimed for packing dimension but the argument works only for Hausdorff dimension as explained in [JJ2].
The study of mean porous measures was pioneered by Beliaev and Smirnov [BS]. In [BS] the proof of the statement that the same upper bound which is valid for packing dimensions of mean porous sets holds for mean porous measures as well is based on a proposition claiming that mean porous measures are approximable by mean porous sets. However, this is not the case: in Theorem 4.1 we construct a mean porous measure such that all mean porous sets have zero -measure. The main purpose of this paper is to develop a new method to show that the statements of [BS, JJ1] are true (see Theorem 3.1) even though the proofs are not correct.
The paper is organized as follows: In Section 2 we discuss the basic concepts. Section 3 is dedicated to the proof of our main result. Besides this, we illustrate by an example that the upper bound which we obtain is asymptotically the best possible one. Finally, in the last section we construct an example of a mean porous measure which is not approximable by mean porous sets.
2. Basic concepts
In this section, we give the basic definitions used throughout the paper. Intuitively, the porosity of a set gives for all small scales the relative radius of the largest ball which fits into a reference ball centered at the set and which does not intersect the set. Let . For all and , we define
[TABLE]
Here is the closed ball with centre at and radius . Clearly, for all . Given , the set is said to be -porous at if
[TABLE]
Moreover, is -porous if it is -porous at every point .
For measures, the corresponding concepts are defined as follows: Let be a Radon measure on . For all and for all positive real numbers and , set
[TABLE]
Given , the measure is -porous at a point if
[TABLE]
The order of taking limits is important here: if we changed it we would obtain the porosity of , the support of . Finally, the measure is -porous if there is with such that is -porous at every point . It is not difficult to see that in this case . For more information on porosity of measures, see [EJJ].
Larger classes of mean porous sets and measures are obtained by demanding that a certain percentage of scales - not necessarily all small ones - are porous. Given and a positive integer , the set is -porous for scale at a point whenever . For , the set is called mean -porous at a point if
[TABLE]
Here the cardinality of a set is denoted by . We say that is mean -porous if it is mean -porous at every point . The measure , in turn, is mean -porous at if
[TABLE]
Finally, is mean -porous if there is with such that is mean -porous at all points .
The packing dimension, , of a Radon measure on is defined in terms of local dimensions as follows:
[TABLE]
where
[TABLE]
and means the essential infimum with respect to . Equivalently, the packing dimension of is given by means of packing dimensions of Borel sets with positive -measure [C]:
[TABLE]
Remark 2.1*.*
Replacing the essential infimum by the essential supremum in the above definition leads to the concept of upper packing dimension. Using the fact that restricting a measure will not decrease the porosity [EJJ], we see that Theorem 3.1 is valid for the upper packing dimension of as well provided that is mean porous -almost everywhere.
3. Packing dimension of measures with large mean porosity
In this section we prove the following packing dimension estimate for mean porous measures:
Theorem 3.1**.**
Let and . There exists a constant depending only on such that for all mean -porous Radon measures on we have
[TABLE]
At the end of this section, we give a construction (Example 3.9) which indicates that the upper bound of Theorem 3.1 is asymptotically the best possible one as tends to . The proof of Theorem 3.1 is given as a series of lemmas. The first one serves as a key tool in the proof of our main result. We use the symbol for the side-length of a cube .
Lemma 3.2**.**
Let and . Let be a Radon measure on with . Assume that all disjoint collections of half-open -adic cubes with side-length at most have the following property: for all there is such that
[TABLE]
Then
[TABLE]
Proof.
Suppose to the contrary that . Then for -almost every . Recall from [C, Lemma 2.3] that for -almost all the local dimension may be calculated using half-open -adic cubes containing instead of balls . Hence, for -almost every we may choose a cube containing and with side-length for some such that . Let be a disjoint collection of such cubes covering -almost all points of . Then
[TABLE]
which is a contradiction. ∎
The following lemma shows that when cubes are of the same size we can approximate the sums from above by distributing the measure evenly on the cubes.
Lemma 3.3**.**
Let . Assume that and are disjoint cubes with side-length . Then for any Radon measure on
[TABLE]
Proof.
Since the cubes are of the same size, the claim follows directly from Hölder’s inequality:
[TABLE]
∎
The proof of the next lemma is based on straightforward geometric arguments. The boundary of a set is denoted by .
Lemma 3.4**.**
Let and let be a dyadic cube. If are closed balls with radii at least , then may be covered by dyadic cubes of side length . Here is a positive and finite constant depending only on .
Proof.
Let and . We may assume that and since . Denote the faces of by , and divide the balls into disjoint sets such that provided is the closest face to . Here the distance is measured from the centre of a face. (If there are several faces which are equally close choose one of those.) Fix and consider the part of determined by . Since we have that if is sufficiently close to . Hence, there is a constant such that the angle between the normal of and the tangent plane of is larger than at any point in for all . The minimum point for this angle is obtained in the following manner: Consider a vertex of contained in . Denote by the vertex of which is opposite to , that is, the line segment determined by and is a diagonal of . Let be the line determined by the diagonal of containing . Move along away from up to the point where the distance to equals . Now is a minimum point which determines . We conclude that there is a bi-Lipschitz injection from to such that the Lipschitz constants depend only on . This gives the claim. ∎
Before stating the rest of the auxiliary results, we introduce the notation we need throughout the remaining part of this section.
Consider the smallest integer such that . For , let be the unique positive integer for which
[TABLE]
For all , the collection of all half-open -adic cubes of side-length is denoted by . Moreover, if and , we use the notation provided that . This is simplified to in the case .
Let be a Radon measure on . Given , we call a cube porous provided that
[TABLE]
for some . For all , set
[TABLE]
Finally, for let be the unique cube containing . If , is the cube obtained from a cube by magnifying by the factor with respect to the centre of .
One of the fundamental and most useful structural properties of porous sets is the following: if the porosity of is close to , then locally inside each ball with radius the set is contained in a small neighbourhood of some -dimensional surface with -measure comparable to . For more precise statements of this type, see e.g. [S] or [JJKS]. In the following lemma which is a slight improvement of [JJ1, Lemma 2.2] we translate this fact into the language of mean porous measures. It states that every cube may be divided into three parts , where has small measure, is a small neighbourhood of some -dimensional surface and contains no porous points.
Lemma 3.5**.**
Let be a Radon measure on , and . Then any cube may be divided into three parts
[TABLE]
where , , and can be covered by at most cubes . Here and are positive and finite constants.
Proof.
For any with , there is such that
[TABLE]
Hence, given with , we find a ball of radius such that
[TABLE]
Denoting by the distance from a point to a set , define
[TABLE]
It is evident from (3.2) that
[TABLE]
On the other hand, since , we have that , and so it follows easily from Lemma 3.4 that may be covered by at most cubes where is the constant of Lemma 3.4.
It remains to show that , that is,
[TABLE]
Consider and choose with . Let and be as in (3.2). By the choice of we have
[TABLE]
which gives , and therefore . This completes the proof of (3.3). ∎
Next lemma is in the core of the proof of Theorem 3.1. It shows that summing over porous subcubes of a cube gives a small factor which is decreasing exponentially under iteration. This is needed when proving that the assumptions of Lemma 3.2 are valid.
For the purpose of formulating our key lemma, we define weights for cubes as follows: Suppose that is a Radon measure on and . Set
[TABLE]
where is as in Lemma 3.5. If for some , define
[TABLE]
For all with , let be the unique real number satisfying
[TABLE]
where for all
[TABLE]
and is as in Lemma 3.5.
Before formulating our lemma we make one more remark: From now on, we assume that is so close to that
[TABLE]
Thus , giving
[TABLE]
for all . This fact will be used repeatedly in the proof of Lemma 3.6.
For any and , we denote by the unique cube in for which . Clearly, and .
Lemma 3.6**.**
Let be a Radon measure on , and with . Then for all and we have
[TABLE]
Moreover, if is any finite collection of disjoint -adic cubes and is a Radon measure on such that , then
[TABLE]
where such that .
Proof.
The proof is based on Lemma 3.5. The problematic part in Lemma 3.5 is the factor . If is close to the boundary of the expanded cube will not be a subset of , and therefore we are unable to estimate the sum in terms of -measure of . This problem is overcome by dividing the subcubes of into two parts depending on there distance to the boundary of . This leads to the use of : the larger we take, the better estimates we have. For we cannot utilize porosity at all.
Claim (3.11) is a direct consequence of repeated applications of (3.10). In what follows we will prove (3.10) by induction. Divide the cube into two regions of subcubes, the boundary region and the interior one , as follows
[TABLE]
and
[TABLE]
For , we estimate the sum in these subcubes in the following manner: Let and be as in (3.6) and (3.7). Take . Defining in terms of , we obtain from definition (3.5) and Lemma 3.3 that
[TABLE]
To estimate the sum over , we apply Lemma 3.5 to . Note that the part is now absent, and therefore the sum is divided into two parts determined by and . Using Lemma 3.3 in both parts and the fact that , we have recalling (3.4)
[TABLE]
Combining (3.12) with (3.13), we continue using Lemma 3.3 and (3.9)
[TABLE]
Recalling that , it is evident that . From (3.9) and Lemma 3.3 we obtain
[TABLE]
Putting together the above estimates, proves (3.10) when .
Next we assume that (3.10) holds when and verify it in the case . Again, we divide into two regions of subcubes defined above. Lemma 3.5 is used to evaluate the sum in the interior region whereas the induction hypothesis is applied in the boundary region . Let and be as in (3.6) and (3.7).
In we begin with
[TABLE]
From (3.12) and (3.13) we get for the inner sum
[TABLE]
Repeating this times and using (3.7), (3.9) and Lemma 3.3 yields
[TABLE]
Notice that for all by (3.7).
Finally, the induction hypothesis, Lemma 3.3 and (3.9) combine to give in the boundary region (recall (3.4))
[TABLE]
The claim follows by summing up (3.14) and (3.15). ∎
In the previous lemma played no role. Our next result quantifies the fact that we really gain something if a large proportion of scales is porous.
Lemma 3.7**.**
Let and
[TABLE]
Suppose that is a Radon measure on . Then there are and such that
[TABLE]
for all and with
[TABLE]
for some . Here and are independent of and .
Proof.
Let
[TABLE]
Choose such that
[TABLE]
Obviously, is porous whenever , and thus
[TABLE]
By (3.5) we have
[TABLE]
which gives the claim by (3.16). ∎
Remark 3.8*.*
Mean porosity is defined in terms of scales where . However, in the proof of Theorem 3.1 we need to use scales of the form where . This problem may be overcome by the following observation: Let and such that . Then there is an integer with such that
[TABLE]
Observe that the starting scale plays no role in Lemmas 3.6 and 3.7 and we choose it (out of possibilities) depending on the point.
Proof of Theorem 3.1.
Suppose that is mean -porous for for which (3.8) holds and for . It is clearly enough to prove the claim for such ’s. We shall prove that
[TABLE]
where and are as in Lemma 3.6. This implies the claim since, by (3.1),
[TABLE]
We may assume without loss of generality that and .
Let , and and be as in Lemma 3.7. Take as in (3.6) with . Since is mean -porous there are a Borel set and constants , and such that and
[TABLE]
for all and for all . Let be so large that
[TABLE]
Let be any finite collection of -adic cubes with for all . For those for which , we define . By Remark 3.8 there are and such that
[TABLE]
provided that . (Note that is added here because it appears in the assumptions of Lemma 3.7 via the way we defined porous cubes in the chapter after (3.1).) For each such fix with and . In this way we obtain collections of -adic cubes scaled by the factor , . Since we have by Lemma 3.7 and (3.18)
[TABLE]
for each where is as in (3.5). Summing over all ’s and using (3.11), we get for all
[TABLE]
Since , inequality (3.19) implies that
[TABLE]
Suppose that with and choose so small that . This choice is possible since as . Now
[TABLE]
Combining (3.20) and (3.21) gives
[TABLE]
Since the same upper bound is valid also for infinite collections of cubes Lemma 3.2 implies that . Letting completes the proof of (3.17). ∎
The following example shows that we cannot get a better upper estimate on the dimension than
[TABLE]
even for mean porous sets. For simplicity, we consider the case . We use the notation for Hausdorff dimension. The idea of the example is the following: First we take steps in the construction of a standard Cantor set where each interval is substituted by two subintervals of length . This will give us scales where porosity is at least . Next we divide each construction interval into subintervals and choose every other of them. This guarantees that the dimension will be large enough but the cost of it is the term in the number of porous scales. To get rid of this we iterate this procedure increasing at every step.
Example 3.9**.**
Fix . Define for all a Cantor-type mean-porous set
[TABLE]
with
[TABLE]
where
[TABLE]
A standard calculation shows that
[TABLE]
On scales , where
[TABLE]
the porosity at each point is at least . Hence, is mean -porous. Finally, we notice that
[TABLE]
4. Mean porous measures are not necessarily approximable by
mean porous sets
The aim of this section is to clarify relations between mean porosities of sets and those of measures by verifying that, contrary to [BS, Proposition 1], mean porous measures cannot be approximated by mean porous sets. For simplicity, we restrict our consideration to . Our goal is to construct a mean porous measure on with the property that any mean porous set has zero -measure.
Intuitively, the measure is constructed in the following way: For all positive integers , we define slowly decaying weights and start with the uniform measure on the unit interval. On step of the construction we redistribute the measure on the dyadic intervals of length by attaching weight to one half of the dyadic interval of length and to the other half. We alternate the half of the interval which gets most of the measure. To be precise, define for all
[TABLE]
and proceed by modifying the construction of [EJJ, Example 4]: For positive integers , set and
[TABLE]
For all binary sequences , we denote by the closed dyadic interval of length whose left endpoint in binary representation is . Let be the unique Radon probability measure on determined by the formula
[TABLE]
for all binary sequences . We change continuously the side of the larger weight since we want to avoid the technical difficulties caused by the fact that at some scales the small ball giving the porosity might be outside the dyadic interval containing the point we are considering.
Theorem 4.1**.**
The measure has the following properties:
- (1)
* for all ,* 2. (2)
* is -porous and* 3. (3)
* is mean -porous for all .*
Remark 4.2*.*
For (1) we need to take slowly decreasing weights whilst for (2) and (3) we only need that as . Note that in (2) the value of porosity is not optimal. Our aim is only to demonstrate that even porous measures need not be approximable by mean porous sets.
Theorem 4.1 will be proven as a consequence of several lemmas. The first one, Lemma 4.3, is a corollary of the strong law of large numbers. For , let denote the :th digit of in the binary representation. The non-uniqueness of the representation plays no role here since is easily seen to be non-atomic.
Lemma 4.3**.**
For -almost all , we have
[TABLE]
Proof.
Applying the strong law of large numbers, see [Fe, Chapter X.7] to the sequence of mutually independent random variables , if is odd, and , if is even, gives
[TABLE]
for -almost all . Here . From this the claim follows easily. ∎
For all positive integers , we use the notation for the half-open dyadic subintervals of having length .(We do not use the notation introduced in Section 3 since we want to distinguish between powers of and .) Letting be a positive integer and , we say that contains an -hole of if there is such that and .
Lemma 4.4**.**
Suppose that is mean -porous for some and . Let be a positive integer such that . For , let be such that . Then for -almost all , we have
[TABLE]
Proof.
Observe that for all there exists such that
[TABLE]
for all . Lemma 4.3 implies that for -almost every we may choose a positive integer with such that
[TABLE]
for all . Hence, for -almost every
[TABLE]
for all .
Consider for which (4.1) holds and a positive integer such that and is -porous for scale at . An easy calculation yields that . Since, by the choice of , there is such that and , it follows that contains an -hole of . Now (4.1) gives the claim. ∎
For the remaining two lemmas we need the following notation: Let and be positive integers and . If and such that , we use the same notation as in Section 3 with replaced by . We say that is porous with respect to if for some . For and , let be the unique interval in for which . Obviously, these definitions depend on the indices and , but in what follows this dependence does not cause any confusion and it is therefore omitted for the sake of simplicity. We refer by to -adic intervals and by to dyadic intervals. In particular for there exists unique intervals for all .
Our next lemma is a consequence of the previous one.
Lemma 4.5**.**
Suppose that is mean -porous for some and and . Let be a positive integer with . Then there are , positive integers and and with such that for all positive integers and for all we have
[TABLE]
Proof.
Lemma 4.4 implies the existence of with and that of a positive integer such that for all positive integers and for all with we have
[TABLE]
It follows that (4.2) is valid for all when . To see this, let . Denote by the largest with . If , then the claim clearly holds by (4.2). On the other hand, assuming that , we obtain
[TABLE]
by the choice of .
Choose a positive integer so large that . Dividing the sequence into successive blocks of length , it follows that for all positive integers and for all we have
[TABLE]
Indeed, if this is not the case, then
[TABLE]
giving a contradiction to (4.2). But if contains an -hole of for some , then is porous with respect to , and we obtain
[TABLE]
for all . ∎
Remark 4.6*.*
The use of and prevents us from counting same holes twice.
Next we introduce for each interval a weight which is equal to 1 if is not porous and 1 minus the portion of the measure of the hole if is porous. For this purpose, let and be positive integers and let . Define
[TABLE]
for all non-negative integers and .
In formula (4.3) of Lemma 4.7 we prove that the product of weights of all predecessors of converges uniformly to 0 as the size of decreases. The reason for this is quite simple: on a positive percentage of scales there are porous intervals with and the product converges to 0 since the measure of the holes decreases very slowly. On the other hand, the proof of formula (4.4) is based on the definition of . It guarantees that we may utilize porous scales.
Lemma 4.7**.**
Suppose that is mean -porous for some and and . Let , and positive integers , and be as in Lemma 4.5. Then for all positive integers and for all we have
[TABLE]
where is a constant depending only on , , and and as . (The dependence on all other parameters but is irrelevant for our purposes.) Moreover,
[TABLE]
for all positive integers .
Proof.
By decreasing , we may assume that is an integer such that . (Note that in Lemma 4.5 the integer is independent of .) Consider . Letting be a positive integer with and applying Lemma 4.5 successively for the indices , , , gives
[TABLE]
implying (4.3).
Inequality (4.4) is valid since
[TABLE]
Here the inequality follows by dividing the inner summation into the two possible cases: either for all or for some . ∎
Now we are ready to prove the main theorem of this section.
Proof of Theorem 4.1.
Assume contrary to the claim (1) that there is and a mean -porous set with . Using Lemmas 4.5 and 4.7, we find positive integers and and with satisfying the estimates (4.3) and (4.4). Taking as in Lemma 4.7, we get for all
[TABLE]
giving as contrary to . This proves (1).
An elementary calculation proves (2), for details see [EJJ, Example 4]. Clearly, is not the best possible value, but this is irrelevant for our purposes.
For the remaining part (3), given , fix a positive integer such that . As in Lemma 4.3, we obtain
[TABLE]
for -almost all . Thus, for , we find for -almost all a positive integer such that
[TABLE]
for all . This implies that is mean -porous at -almost all points (for details, see [EJJ, Example 4]), and consequently (3) holds. ∎
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[BS] D. B. Beliaev and S. K. Smirnov, On dimension of porous measures , Math. Ann. 323 (2002), 123–141.
- 2[C] C. D. Cutler, Strong and weak duality principles for fractal dimension in Euclidean space , Math. Proc. Cambridge Philos. Soc. 118 (1995), 393–410.
- 3[EJJ] J.-P. Eckmann, E. Järvenpää and M. Järvenpää, Porosities and dimensions of measures , Nonlinearity 13 (2000), 1–18.
- 4[Fe] W. Feller, An Introduction to Probability Theory and Its Applications , John Wiley & Sons, New York, 1950.
- 5[JJ 1] E. Järvenpää and M. Järvenpää, Porous measures on ℝ n superscript ℝ 𝑛 \mathbb{R}^{n} : local structure and dimensional properties , Proc. Amer. Math. Soc (2) 130 (2002), 419–426.
- 6[JJ 2] E. Järvenpää and M. Järvenpää, Average homogeneity and dimensions of measures , Math. Ann. 331 (2005), 557–576.
- 7[JJKS] E. Järvenpää, M. Järvenpää, A. Käenmäki and V. Suomala, Asymptotically sharp dimension estimates for k 𝑘 k -porous sets , Math. Scand. 97 (2005), 309–318.
- 8[KR] P. Koskela and S. Rohde, Hausdorff dimension and mean porosity , Math. Ann. 309 (1997), 593–609.
