Differentiating along rectangles with fixed shapes in a set of directions
Emma D'Aniello, Laurent Moonens

TL;DR
This paper investigates how certain shape-dependent rectangles can be used to differentiate functions in specific Orlicz spaces, revealing conditions under which differentiation is possible or not.
Contribution
It introduces a shape-function framework for rectangles in the plane and analyzes their ability to differentiate various Orlicz spaces, including new examples and limitations.
Findings
Certain shape-functions enable differentiation of L log L spaces.
Fast-growing shape-functions may fail to differentiate L log^α L spaces.
Examples demonstrate the impact of shape-function growth on differentiation capabilities.
Abstract
In the present note, we examine the behavior of some homo\-thecy-invariant differentiation basis of rectangles in the plane satisfying the following requirement: for a given rectangle to belong to the basis, the ratio of the largest of its side-lengths by the smallest one (which one calls its \emph{shape}) has to be a fixed real number depending on the angle between its longest side and the horizontal line (yielding a \emph{shape-function}). Depending on the allowed angles and the corresponding shape-function, a basis may differentiate various Orlicz spaces. We here give some examples of shape-functions so that the corresponding basis differentiates , and show that in some `model' situations, a fast-growing shape function (whose speed of growth depends on ) does not allow the differentiation of .
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.
Taxonomy
TopicsAdvanced Mathematical Modeling in Engineering · Advanced Numerical Analysis Techniques · Point processes and geometric inequalities
Differentiating along rectangles
with fixed shapes in a set of directions
Emma D’Aniello and Laurent Moonens
(Date: March 17, 2024)
Abstract.
In the present note, we examine the behavior of some homothecy-invariant differentiation basis of rectangles in the plane satisfying the following requirement: for a given rectangle to belong to the basis, the ratio of the largest of its side-lengths by the smallest one (which one calls its shape) has to be a fixed real number depending on the angle between its longest side and the horizontal line (yielding a shape-function). Depending on the allowed angles and the corresponding shape-function, a basis may differentiate various Orlicz spaces. We here give some examples of shape-functions so that the corresponding basis differentiates , and show that in some “model” situations, a fast-growing shape function (whose speed of growth depends on ) does not allow the differentiation of .
Key words and phrases:
Maximal functions; differentiation bases.
2010 Mathematics Subject Classification:
Primary: 42B25, 28B05; Secondary: 28D05.
Laurent Moonens acknowledges the partial support of the “Luigi Vanvitelli” University of Campania, through a scholarship awarded to him as a “Visiting Professor”.
A (Buseman-Feller) differentiation basis in the plane is a collection of open sets such that for every one has . In the sequel, we shall always assume that is homothecy-invariant, meaning that one has for all and all .
It can arise, given a locally integrable function , that Lebesgue’s differentiation theorem’s conclusion holds for using sets in instead of the usual Euclidean balls, namely that one has, for a.e. :
[TABLE]
In case the latter holds, we shall say that differentiates . If differentiates for all , we shall than say, for simplicity, that differentiates . If a basis differentiates , we call it a density basis.
It is well-known (see e.g. De Guzmán’s book [5]) that, for a homothecy-invariant basis as above, the two following properties are equivalent for a given Young function (one ofter refers to this as the Sawyer-Stein principle for differentiation bases)
- (i)
differentiates the Orlicz space ;
- (ii)
the maximal operator defined by satisfies the following estimate for all measurable and all :
[TABLE]
When is a homothecy-invariant collection of rectangles, various situations can occur; here are some famous examples:
- •
if is the collection of all rectangles parallel to the axes, then is the largest Orlicz space that differentiates (see Stokolos [10]);
- •
if is the collection of all rectangles (parallel to the axes or not), there always exists such that is not differentiated by (see Buseman and Feller [2]) — actually the same conclusion holds even if one replaces the collection of all rectangles by the collection of rectangles one side of which makes an angle with the horizontal line belonging to some countable sets, like e.g. any set that is dense in some interval (as it follows from [2]) or even (see De Guzmán [6]);
- •
if, though, is the set of all rectangles one side of which makes an angle with the horizontal line belonging to the set (or to the image of any lacunary sequence as defined in [8]), then is known to differentiate for all (see Córdoba and Fefferman [3] for and Nagel, Stein and Wainger [9] for all );
- •
it actually follows from a beautiful paper by Bateman [1] that if is a given set of angles, then the basis of all rectangles one side of which makes an angle with the horizontal line, either differentiates all for all , or fails to differentiate any for (dichotomy which, as we observed with J.M. Rosenblatt in [4], remains true if one replaces the range by );
- •
it also follows from [1] that never differentiates any for (and hence never is a density basis according to [4]) if is uncountable.
Defining the shape of a rectangle as the quotient of its longest side-length by its shortest side-length, positive or negative differentiation results can also depend on restrictions made on this ratio. Let us mention, for example, a few situations where this influence is well understood:
- •
if is a homothecy-invariant basis of rectangles whose shapes are bounded from above, then differentiates (this easily follows from the fact that the maximal operator then behaves distributionally like the uncentered Hardy-Littlewood maximal operator on balls);
- •
if for any , denotes the set of all ternary numbers of the form for some , (which one can see as a “truncated ternary Cantor set”), then the basis of all rectangles such that, for some , has shape and has its longest side making an angle satisfying , fails to differentiate for all , as it follows from Katz [7].
In this short note, we focus our attention on the following question: given an infinite set of angles such that [math] is a limit point of , does there exist a function satisfying , for which the basis differentiates (or fails to differentiates) a given Orlicz space, where denotes the basis of all rectangles for which there exists so that the longest side of makes an angle with the horizontal line, and that ? We here examinate the following situations:
- •
under the above assumptions, the basis never differentiates as it follows from a result by Moriyón (see Proposition 1 below);
- •
given small enough and any set , there always exists a (nonincreasing) function satisfying the above conditions and for which differentiates exactly (see Corollary 3, which basically follows from a simple geometric observation we describe in section 2, and from a result by Stokolos [10]);
- •
if is obtained from the “model” geometrical sequence by inserting uniformly angles in between and , and if is constant on each of those “blocks”, then, depending on how , and the Young function behave with respect to each other, it may happen that the associated basis fails to differentiate the Orlicz space (recall that it is still an open problem whether does or does not differentiate for ); here of course could be replaced by any lacunary sequence in the sense of [8].
1. Using a result by R. Moriyón
We keep the notations defined in the introduction.
Proposition 1**.**
Assume that and satisfy . In this case, the homothecy-invariant basis fails to differentiate .
Proof.
Define, as in Moriyón’s theorem (cited in [5, Appendix III, p. 206]), the set:
[TABLE]
Using the fact that , choose for which . Define then as the rectangle with area centered at the origin, having shape and its longest side making an angle with the horizontal line, so that one has , , and hence also . The length of its longest side being equal to , which can be arbitrary large, it is clear that is unbounded. It hence follows from condition e) in Moriyón’s theorem (cited in [5, Appendix III, p. 206]) that fails to differentiate . ∎
2. From a simple geometrical observation to bases differentiating
2.1. A simple geometrical observation
Fix a rectangle with longest side and shortest side , such that its longest side makes an angle with the origin. Fix then a parameter and denote by the rectangle parallel to the axes contained inside and determined by the fact that two opposite vertices meet the pair of longest sides of at points distant of from the nearest vertex of lying on the same side (see Figure 1 below). Denote by and the horizontal and vertical sides of . On the other hand, denote by the smallest rectangle parallel to the axes containing , and call and its horizontal and vertical sides, respectively.
Simple trigonometric computations yield:
[TABLE]
[TABLE]
[TABLE]
and:
[TABLE]
Letting denote the shape of , it’s now a routine computation to calculate:
[TABLE]
and:
[TABLE]
Define now the ratio:
[TABLE]
Assuming that is small enough, observe that one has (so that the construction makes sense) provided that one has:
[TABLE]
On the other hand, one computes:
[TABLE]
so that in the above range for , the function is increasing if is fixed (and small enough) and tends to as approaches .
Finally, one gets also, for :
[TABLE]
so that is also an increasing map in the latter range for .
Now fix a set of which [math] is a limit point, fix a large number and choose, for any , a ratio such that one has:
[TABLE]
note that this is possible since we have, for all :
[TABLE]
Observe, using what has been said before, that is decreasing. Computing moreover the value of in terms of (and ), we get easily:
[TABLE]
which yields , .
2.2. Constructing a differentiation basis
Given for which [math] is a limit point, we constructed in the previous section a function . Let now be the basis, defined in the introduction, of all rectangles for which there exists a so that and that the longest side of makes an angle with the horizontal line. Define also two basis of two-dimensional intervals by:
[TABLE]
Lemma 2**.**
Given a measurable function , one has:
[TABLE]
Proof.
Assume first that is given and compute, using the previous notations:
[TABLE]
We hence have:
[TABLE]
If now one fixes , then denote by a rectangle satisfying . One then writes:
[TABLE]
and we hence get . ∎
Corollary 3**.**
The basis differentiates exactly .
To prove this corollary, we shall need the following lemma, relying mainly on [10].
Lemma 4**.**
Assume that is a homothecy-invariant, Buseman-Feller differentiation basis of rectangles parallel to the coordinate axes in . If , then differentiates exactly .
Proof of the lemma.
To prove this lemma, observe that one can, without loss of generality, assume that all rectangles in have their longest side parallel to the -axis. It is clear indeed, that one can write , where and are the collection of elements of whose longest side lie in the - and - direction, respectively. One has then for at least one . It then follows from symmetrization if necessary, than one cas always assume . Now just note that if differentiates exactly , then the same is true for .
As in [10], denote by the set of all dyadic parents of elements in (the dyadic parent of a rectangle parallel to the axes, being the rectangle with dyadic side-lenths containing , concentric with it and having the smallest possible area). Since it is easy to check that one has for any rectangle parallel to the axes, is now clear that one has . It is also easy to see that is translation invariant. Finally, observe that is also invariant under homothecies with dyadic ratio. Indeed, fix and and let’s see that . We need to establish that for some . Yet if one denotes by and (resp. and ) the lenths of the - and -sides of (resp. ) respectively, we get by definition of the dyadic parent and . This also yields and , meaning that has side-lengths and . Denoting by the rectangle parallel to the axes with same center as and side-lengths and respectively, it is then clear that is homothetic to (and so that one has ) while one has . Hence , what we wanted to prove.
Now take a strictly increasing sequence verifying and for which is a strictly increasing sequence.
Let for all . Observing first that , translation-invariance of ensures that one has . By the preceding comments, it hence follows that one has .
Since now, for all , the family is a finite subset of with pairwise incomparable elements up to translation (writing , it is clear indeed that the -side length of increases with , while its -side length decreases), it follows that enjoys property (S) of [10], and hence differentiates exactly . ∎
Proof of Corollary 3.
Given a rectangle whose longest side makes an angle with the horizontal axis, and satisfying , we compute using the previous results:
[TABLE]
and:
[TABLE]
Since both of those ratios tend to when approaches [math], it follows from Lemma 4 that and differentiate exactly . Now use Lemma 2 to infer, by the Sawyer-Stein principle (see (1) in the introduction), that differentiates exactly . ∎
Remark 5**.**
A simple computations shows that the shape-function constructed before in such a way that differentiates , has a linear growth with respect to (meaning that there are constants for which one has for all small enough).
3. Shape-functions constant on blocks
We now examine the case where
[TABLE]
is associated to a sequence decreasing to [math] and to a sequence of integers by inserting uniformly angles in between and , and where is constant on each of those “blocks”, meaning that for each , there exists a real number such that one has for all . We remain in this setting until the end of this section, unless otherwise mentioned.
Recall that, given sequences and , one writes (resp. ) if the quotient tends to [math] (resp. remains bounded) as grows to .
Proposition 6**.**
Assume and are as before and that one has moreover, for each :
[TABLE]
Let be a Young function. If the homothecy-invariant basis associated to differentiates the Orlicz space , then one has as .
Proof.
We follow a similar strategy to the one developed by the second author in [8]. To this purpose, define for each a two-dimensional interval . Define, for , to be the rectangle obtained from by rotating it around the origin by an angle . It is not hard to observe that condition (2) ensures that one has:
[TABLE]
by noting for example that (2) implies that (at least) half (in area) of the rectangle has no overlap with for . Define then and let be the set obtained by rotating around the origin by an angle (see Figure 2).
Given and , there is an such that one has . But since one has , this yields:
[TABLE]
We hence have . Yet if did differentiate , it would follow from (1) that one would have:
[TABLE]
from which the announced statement follows. ∎
The previous result has amusing consequences when one considers sets of the previous form associated to the sequence defined by (recall from the introduction that for this sequence it is known that differentiates for all , with no restriction on the shapes of the rectangles, while it is unknown if it differentiates ). The corollary below shows that adding angles in between two terms of the “model” geometric sequence, while restricting the shape in those “blocks”, may fail to differentiate Orlicz spaces lying in between and for all — of course, the shape-function therefore has to increase quite fast.
Corollary 7**.**
Assume that for all and let be associated to and to as before. Define also, for , and let be the associated shape-function. If, for some , one has when , then fails to differentiate .
Proof.
This corollary follows in a straightforward way by contradiction from the previous proposition applied to . ∎
Remark 8**.**
One can of course formulate a similar corollary, starting from a lacunary sequence as in [8] instead of the “model” sequence .
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] Bateman, M. Kakeya sets and directional maximal operators in the plane. Duke Math. J. 147 (2009), no. 1, 55–77.
- 2[2] Busemann, H. and Feller, W. Zur Differentiation der Lebesgueschen Integrale. Fundam. Math. 22 (1934), 226–256.
- 3[3] Córdoba, A.; Fefferman, R. On differentiation of integrals. Proc. Nat. Acad. Sci. U.S.A. 74 (1977), no. 6, 2211–2213.
- 4[4] D’Aniello, E.; Moonens, L.; Rosenblatt, J. Differentiating Orlicz spaces with rare bases of rectangles Submitted , ar Xiv:1808.07283.
- 5[5] de Guzmán, M. Differentiation of Integrals in ℝ n superscript ℝ 𝑛 {\mathbb{R}}^{n} . Lecture Notes in Mathematics. Springer-Verlag 481 , 1975.
- 6[6] de Guzmán, M. Real Variable Methods in Fourier Analysis. Mathematics Studies 46 , North-Holland, 1981.
- 7[7] Katz, N.H. A counterexample for maximal operators over a Cantor set of directions. Math. Res. Lett. 3 (1996), no. 4, 527–536.
- 8[8] Moonens, L. Differentiating along rectangles, in lacunary directions. New York J. Math. 22 (2016), 933–942.
