Exact order of extreme \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_p$$\end{document}Lp discrepancy of infinite sequences in arbitrary dimension

We study the extreme \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_p$$\end{document}Lp discrepancy of infinite sequences in the d-dimensional unit cube, which uses arbitrary sub-intervals of the unit cube as test sets. This is in contrast to the classical star \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_p$$\end{document}Lp discrepancy, which uses exclusively intervals that are anchored in the origin as test sets. We show that for any dimension d and any \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p>1$$\end{document}p>1, the extreme \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_p$$\end{document}Lp discrepancy of every infinite sequence in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[0,1)^d$$\end{document}[0,1)d is at least of order of magnitude \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\log N)^{d/2}$$\end{document}(logN)d/2, where N is the number of considered initial terms of the sequence. For \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p \in (1,\infty )$$\end{document}p∈(1,∞), this order of magnitude is best possible.

where λ denotes the Lebesgue measure of B. A global discrepancy measure is then obtained by considering a norm of the local discrepancy with respect to a fixed class of test sets.
The first author is supported by the Austrian Science Fund (FWF), Project F5509-N26, which is a part of the Special Research Program "Quasi-Monte Carlo Methods: Theory and Applications".
In the following, let p ∈ [1, ∞). The classical (star) L p discrepancy uses as test sets the class of axis-parallel rectangles contained in the unit cube that are anchored in the origin. The formal definition is The extreme L p discrepancy uses as test sets arbitrary axis-parallel rectangles contained in the unit cube. For u = (u 1 , u 2 , . . . , u d ) and The extreme L p discrepancy of P is then defined as Note that the only difference between the standard and the extreme L p discrepancy is the use of anchored and arbitrary rectangles in Of course, with the usual adaptions the above definitions can be extended also to the case p = ∞. However, it is well known that in this case the star and extreme L ∞ discrepancies are equivalent in the sense that L star ∞,N (P) ≤ L extr ∞,N (P) ≤ 2 d L star ∞,N (P) for every N -element point set P in [0, 1) d . For this reason, we restrict the following discussion to the case of finite p.
Recently it has been shown that the extreme L p discrepancy is dominated -up to a multiplicative factor that depends only on p and d -by the star L p discrepancy (see [21,Corollary 5]), i.e., for every d ∈ N and p ∈ [1, ∞), there exists a positive quantity c d,p such that for every N ∈ N and every N -element point set in [0, 1) d , we have For p = 2, we even have c d,2 = 1 for all d ∈ N; see [17,Theorem 5]. A corresponding estimate the other way round is in general not possible (see [17,Section 3]). So, in general, the star and the extreme L p discrepancy for p < ∞ are not equivalent, which is in contrast to the L ∞ -case. For the star L p discrepancy, the result for p ≥ 2 is a celebrated result by Roth [29] from 1954 that was extended later by Schmidt [31] to the case p ∈ (1, 2). For the extreme L p discrepancy, the result for p ≥ 2 was first given in [17,Theorem 6] and for p ∈ (1, 2), in [21]. Halász [15] for the star discrepancy and the authors [21] for the extreme discrepancy proved that the lower bound is even true for p = 1 and d = 2, i.e., there exists a positive number c 1,2 with the following property: for every N -element P in [0, 1) 2 with N ≥ 2, we have On the other hand, it is known that for every d, N ∈ N and every p ∈ [1, ∞), there exist N -element point sets P in [0, 1) d such that Possible implied dependencies of C are indicated as lower indices of the symbol.) Due to (1), the upper bound (3) even applies to the extreme L p discrepancy. Hence, for p ∈ (1, ∞) and arbitrary d ∈ N (and also for p = 1 and d = 2), we have matching lower and upper bounds. The result in (3) was proved by Davenport [6] for p = 2, d = 2, by Roth [30] for p = 2 and arbitrary d and finally by Chen [3] in the general case. Other proofs were found by Frolov [14], Chen [4], Dobrovol'skiȋ [12], Skriganov [32,33], Hickernell and Yue [16], and Dick and Pillichshammer [9]. For more details on the early history of the subject, see the monograph [1]. Apart from Davenport, who gave an explicit construction in dimension d = 2, these results are pure existence results and explicit constructions of point sets were not known until the beginning of this millennium. First explicit constructions of point sets with optimal order of star L 2 discrepancy in arbitrary dimensions have been provided in 2002 by Chen and Skriganov [5] for p = 2 and in 2006 by Skriganov [34] for general p. Other explicit constructions are due to Dick and Pillichshammer [11] for p = 2, and Dick [7] and Markhasin [23] for general p. Bounds for infinite sequences. For the star L p discrepancy, the situation is also more or less clear. Using a method from Proȋnov [25] (see also [10]), the results about lower bounds on star L p discrepancy for finite sequences can be transferred to the following lower bounds for infinite sequences: for every For d = 1, the result holds also for the case p = 1, i.e., for every S in [0, 1), we have L star 1,N (S) ≥ C 1,1 log N for infinitely many N ∈ N. Matching upper bounds on the star L p discrepancy of infinite sequences have been shown in [11] (for p = 2) and in [8] . So far, the extreme L p discrepancy of infinite sequences has not yet been studied. Obviously, due to (1), the upper bounds on the star L p discrepancy also apply to the extreme L p discrepancy. However, a similar reasoning for obtaining a lower bound is not possible. In this paper, we show that the lower bound (4) also holds true for the extreme L p discrepancy. Thereby we prove that for fixed dimension d and for p ∈ (1, ∞), the minimal extreme L p discrepancy is, like the star L p discrepancy, of exact order of magnitude (log N ) d/2 when N tends to infinity.

The result.
We extend the lower bound (4) for the star L p discrepancy of infinite sequences to extreme L p discrepancy.
For d = 1, the results even holds true for the case p = 1.
For the proof, we require the following lemma. For the usual star discrepancy, this lemma goes back to Roth [29]. We require a similar result for the extreme discrepancy. Consider a sub-interval of the (d+1)-dimensional unit cube of the form Vol. 118 (2022) Exact order of extreme Lp discrepancy of infinite sequences 173 and therefore Also ⎛ ⎜ ⎝ Proof of Theorem 1. We use the notation from Lemma 2. For the extreme L p discrepancy of the finite point set P N,d+1 in [0, 1) d+1 , we obtain from [21, Corollary 4] (for d ∈ N and p > 1) and [21,Theorem 7] (for d = 1 and p = 1) that for some real c d+1,q > 0 which is independent of N . Let α d,p ∈ (0, 2 1 p −1 c d+1,p ) and let N ∈ N be large enough such that Vol. 118 (2022) Exact order of extreme Lp discrepancy of infinite sequences 175 According to Lemma 2, there exists an n ∈ {1, 2, . . . , N} such that Thus we have shown that for every large enough N ∈ N, there exists an n ∈ {1, 2, . . . , N} such that It remains to show that (6) holds for infinitely many n ∈ N. Assume on the contrary that (6) holds for finitely many n ∈ N only and let m be the largest integer with this property. Then choose N ∈ N large enough such that For this N , we can find an n ∈ {1, 2, . . . , N} for which (5) and (6) hold true. However, (5) implies that n > m which leads to a contradiction since m is the largest integer such that (6) is true. Thus we have shown that (6) holds for infinitely many n ∈ N and this completes the proof.
As already mentioned, there exist explicit constructions of infinite sequences S d in [0, 1) d with the property that  (7) implies that the lower bound from Theorem 1 is best possible in the order of magnitude in N for fixed dimension d.

Remark 3.
Although the optimality of the lower bound in Theorem 1 is shown by means of matching upper bounds on the star L p discrepancy, we point out that in general the extreme L p discrepancy is really lower than the star L p discrepancy. An easy example is the van der Corput sequence S vdC in dimension d = 1, whose extreme L p discrepancy is of the optimal order of magnitude but its star L p discrepancy is only of order of magnitude log N since lim sup For a proof of (9), see, e.g., [2,27] for p = 2 and [24] for general p. A proof of (8) can be given by means of a Haar series representation of the extreme L p discrepancy as given in [ N elements of the van der Corput sequence, but these can be found in [20]. Employing these estimates yields after some lines of algebra the optimal order result (8).

Remark 4.
The periodic L p discrepancy is another type of discrepancy that is based on the class of periodic intervals modulo one as test sets; see [17,21]. Denote it by L per p,N . The periodic L p discrepancy dominates the extreme L p discrepancy because the range of integration in the definition of the extreme L p discrepancy is a subset of the range of integration in the definition of the periodic L p discrepancy, as already noted in [17,Eq. (1)] for the special case p = 2. Furthermore, it is well known that the periodic L 2 discrepancy, normalized by the number of elements of the point set, is equivalent to the diaphony, which was introduced by Zinterhof [35] and which is yet another quantitative measure for the irregularity of distribution; see [22,Theorem 1] or [18, p. 390] |h j |). Now, for every p > 1 for every infinite sequence S d in [0, 1) d , we have for infinitely many N ∈ N the lower bound Choosing p = 2, we obtain Thus, there exists a positive C d such that for every sequence S d in [0, 1) d , we have This result was first shown by Proȋnov [26] by means of a different reasoning. The publication [26] is only available in Bulgarian; a survey presenting the relevant result is published by Kirk [19]. At least in dimension d = 1, the lower bound (10) is best possible since, for example, F N (S vdC ) √ log N/N for all N ∈ N \ {1} as shown in [28] (see also [2,13]). Note that this also yields another proof of (8) for the case p = 2. A corresponding result for dimensions d > 1 is yet missing.
Funding Open access funding provided by Johannes Kepler University Linz.
Vol. 118 (2022) Exact order of extreme Lp discrepancy of infinite sequences 177 Open Access. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons. org/licenses/by/4.0/.
Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.