Universally bad integers and the 2-adics (Q1876234)

Let \(\mathbb S=\{0,1,4,5,16,17,20,21,\dots\}\) be the set of integers constructed by taking sums of finite subsets of the even powers of 2 (0 corresponding to the empty set). One readily verifies that every integer has a unique representation of the form \(s_1-2s_2\), \(s_1,s_2\in\mathbb S\), and we write \(\mathbb Z=\mathbb S\ominus2\mathbb S\). There are other ways of representing \(\mathbb Z\) as a direct difference of multiples of \(\mathbb S\), e.g., \(\mathbb Z=\mathbb S\ominus4\mathbb S\) and \(\mathbb Z=7\mathbb S\ominus26\mathbb S\). De Bruijn raised the question of classifying pairs of positive odd integers \((a,b)\) such that \(\mathbb Z=a\mathbb S\ominus2b\mathbb S\); he called \((a,b)\) ``good'' in this case and ``bad'' otherwise. The paper under review tackles this problem by means of 2-adic integers. Letting \(\mathbb Z_2\) denote the 2-adic integers and letting \(\overline S\) denote the closure of \(S\subset \mathbb Z_2\) in the 2-adic topology, it is shown that for any pair of positive odd integers \((a,b)\) we have \[ \mathbb Z_2=\overline{a\mathbb S\ominus2b\mathbb S} =\overline{a\mathbb S}\ominus\overline{2b\mathbb S}. \] From this it immediately follows that \((a,b)\) is bad if and only if there is an integer \(n\) such that \(n=a\sigma-2b\tau\), where \(\sigma\) and \(\tau\) are in \(\overline{\mathbb S}\) but not in \(\mathbb S\). Using this characterization the authors construct several classes of bad pairs \((a,b)\).