The Halton sequence and its discrepancy in the Cantor expansion (Q682115)

From MaRDI portal
Revision as of 02:48, 15 July 2024 by ReferenceBot (talk | contribs) (‎Changed an Item)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
scientific article
Language Label Description Also known as
English
The Halton sequence and its discrepancy in the Cantor expansion
scientific article

    Statements

    The Halton sequence and its discrepancy in the Cantor expansion (English)
    0 references
    0 references
    0 references
    0 references
    13 February 2018
    0 references
    For \(s\) bounded sequences \(\underline b_i=(b_{i,j})_{j\in\mathbb N}\) of naturals \(b_{i,j}>1\) with \(b_{i,j},b_{k,l}\) coprime for \(i\neq k\) the generalized Halton sequence \((\omega(n))_{n\in\mathbb N}=((\omega_1,\ldots,\omega_n))_{n\in\mathbb N}\) in \([0,1]^s\) is defined by \(\omega_i(n)=\sum_{l=1}^{L_i(n)}\frac{n_{i,l}}{b_{i,1}\cdot b_{i,l}}\) where \(n_{i,l}\) is the \(l-\)th coefficient of the \(\underline b_i\)-adic expansion of \(n\): \(n=n_{i,1}+n_{i,2}b_{i,1}+n_{i,3}b_{i,1}b_{i,2}+\ldots n_{i,L_i}b_{i,1}\cdots b_{i,L_i-1}\), with \(n_{i,l}\in\{0,1,\ldots,b_{i,l}-1\}\). For \(s=1\) this is the generalized van der Corput sequence to base \(\underline b\), for constant sequences \((b_{i,j})_{j\in\mathbb N}=(b_{i})_{j\in\mathbb N}\) this gives the classical Halton sequence, which is known to be a low-discrepancy sequence, i.e. \(ND^*_N(\omega)=O(\log(N)^s)\), \(D^*_N(\omega)\) denotes the *-discrepancay of the Halton sequence. The main result states that generalized Halton sequences as defined above are also low-discrepancy sequences. The proof is based on techniques developed by \textit{E. Atanassov} [Math. Balk., New Ser. 18, No. 1--2, 15--32 (2008; Zbl 1088.11058)]. A further result states that for \(s=1\) the general Halton sequence \((\omega(n))_{n\in\mathbb N}\) to base \(\underline b=(b_j)_{j\in\mathbb N}\) is always a low-discrepancy sequence even if \((b_j)_{j\in\mathbb N}\) is unbounded. This result however is based on the faulty assumption that partitioning the first \(N\) naturals similar to the binary case yields arithmetic progressions as in the binary case. However it was shown by \textit{H. Chaix and H. Faure} [Acta Arith. 63, No. 2, 103--141 (1993; Zbl 0772.11022)] (Théorème 4.7 for \(p=\infty\)) that in the one-dimensional case \(s=1\) the sequence \((\omega(n))_{n\in\mathbb N}\) to base \(\underline b\) is a low-discrepancy sequence if and only if \(\sum_{i=1}^nb_i=O(n)\).
    0 references
    Halton sequence
    0 references
    Hammersley point set
    0 references
    low discrepancy sequence
    0 references
    pseudo random numbers
    0 references
    generalized van der Corput sequence
    0 references
    Atanassov's method
    0 references

    Identifiers