Probabilistic Diophantine approximation and the distribution of Halton-Kronecker sequences (Q2252046): Difference between revisions

The discrepancy of the first \(N\) elements of a sequence \(z=(z_n)\subseteq [0,1)^d\) is defined by \[ D_N =\sup\limits_B \bigg| \frac{1}{N} \# \{ n\in [0,N): z_n \in B\} -\lambda (B)\bigg| , \] where \(\lambda\) denotes the Lebesgue measure in \([0,1)^d\) and the sup is taken over all rectangular boxes in \([0,1)^d\). It has been conjectured for a long time that \(D_N \geq c_d \frac{(\log N)^d}{N}\) for a constant \(c_d>0\) and for infinitely many \(N\). A \(s\)-dimensional Halton sequence \(x_n=(x_n^{(1)}, \ldots,x_n^{(s)})\) is defined by \(s\) pairwise coprime positive integers \(b_1,\ldots,b_s\), as \(x_n^{(j)}=\sum_{k=0}^\infty m_k b_j^{-k-1}\) if \(n=\sum_{k=0}^\infty m_k b_j^k\) is the customary base \(b_j\) representation. A \(t\)-dimensional Kronecker sequence \(y_n =y_n ({\mathbf \alpha})=(y_n^{(1)},\ldots,y_n^{(t)})\) is defined by a \(t\)-tuple \({\mathbf \alpha}=(\alpha_1,\ldots,\alpha_t) \in [0,1)^t\) as \(y_n^{(j)}=\{ n\alpha_j\}\). The main result of this paper is that, for every \(s\)-dimensional Halton sequence \((x_n)\) and almost every \({\mathbf \alpha}=(\alpha_1,\ldots,\alpha_t)\in [0,1)^t\), the discrepancy of the \((s+t)\)-dimensional Halton-Kronecker sequence \((z_n)=(x_n,y_n({\mathbf \alpha}))\) is low, in the sense that \[ D_N =O\bigg( \frac{(\log N)^{s+t+\varepsilon}}{N}\bigg) \] for every \(\varepsilon >0\). Previously, \textit{R. Hofer} and \textit{G. Larcher} [Math. Z. 271, No. 1--2, 1--11 (2012; Zbl 1257.11073)] proved this result for \(t=1\), while in general they only established the bound above with the larger exponent \(s+t+1+\varepsilon\) instead of \(s+t+\varepsilon\). The current (possibly optimal) improvement was obtained by systematic use of the techniques from the ground-breaking work by \textit{J. Beck} [Ann. Math. (2) 140, No. 2, 451--502 (1994; Zbl 0820.11045)].