Lower bounds for wrap-around \(L_2\)-discrepancy and constructions of symmetrical uniform designs (Q2576280)

The main purpose of this paper is to present explicit lower bounds for measuring the uniformity of symmetric U-type designs (i.e. uniform designs restricted to balanced lattice designs) with any level, under the wrap-around \(L_2\) discrepancy (abbreviated \(WD_2\)). \(WD_2\) is used as the benchmark of uniformity to construct uniform designs via a new class of combinatorial configurations called perfect resolvable balanced incomplete block designs (PRBIBD). Furthermore, an efficient balance-pursuit heuristic for generating any symmetrical uniform design is proposed. The first, introductory Section exposes the basic notions and problems. Section 2 comprises lower bounds of \(WD_2\) on U-type uniform designs, and gives the corresponding sufficient conditions for an U-type design to achieve the lower bound. Based on the obtained sufficient conditions, Section 3 describes two methods, combinatorial and optimization approaches, for constructing uniform designs or low-discrepancy designs under \(WD_2\). The concept of PRBIBD, the connection between uniform designs and PRBIBDs, and certain results on the existence of infinite classes for PRBIBDs are analyzed. The basic techniques included in these methods may also be paralleled to deal with uniform designs under discrepancies other than \(WD_2\). Namely, the existing or newly defined combinatorial configurations can be utilized to construct uniform designs under several different discrepancies, e.g. centered \(L_2\)-discrepancy or other ones. This is a suitable tool to establish not only a relationship between combinatorial designs and uniform designs, but also to provide a platform for unifying different discrepancies by using various properties of combinatorial configurations. Section 3 introduces also an efficient algorithm, named balance-pursuit heuristic, to search for new uniform designs under \(WD_2\), especially with high-level ones. Several tests show that the new algorithm is more powerful than other threshold accepting ones in the literature. \(WD_2\) has been used in quasi-Monte Carlo methods, especially in experimental design, with many applications of uniform design in manufacturing, system engineering, pharmaceutics and natural sciences.