Neighboring powers (Q2268802)

Let \(m\) <and \(n\) be coprime integers with \(n>m\geq 2\). It follows from ten \(abc\)--conjecture that, for each \(\varepsilon >0\) there exists \(c(\varepsilon)>0\) such that for any positive integers \(x\), \(y\) with \(x^m\not=y^n\) we have \[ |x^m-y^n| > c(\varepsilon)X^{1-1/m-1/n- \varepsilon}, \quad X=\max\{x^m,y^n\}. \] An effective much weaker result is a consequence of Baker's theory of linear forms in logarithms: \[ |x^m-y^n| \gg (\log X)^\delta, \] where \(\delta\) is a positive number which depends on \(m\) and \(n\). The authors propose the following conjecture. \textbf{Conjecture.} Let \(m\) <and \(n\) be coprime integers with \(n>m\geq 2\). Then, for any \(c>0\) there exist infinitely many positive integers \(x\), \(y\) such that \[ 0 < |x^m-y^n|< c X^{1-1/m-1/n}, \quad X=\max\{x^m,y^n\}. \] The paper contains weak variants of this conjecture, whose proofs depend upon polynomial constructions. The authors also study many special cases of \((m,n)\). From the text of the addendum: Yann Bugeaud has drawn the authors' attention to the paper by \textit{L. V. Danilov} [Math. Notes 46, No. 5, 914--919 (1989); translation from Mat. Zametki 46, No. 6, 38--45 (1989; Zbl 0694.10020)]. In his article Danilov proves three results which may be shown to be equivalent to three results from our paper. In particular his Theorems 4, 5 and 7 are equivalent to Theorems 1.3, 1.5 and 1.7 respectively in our paper. We remark that the proof of Theorem 6 of Danilov's paper yields a result which is equivalent to the result \(\gamma(9, 11)\leq 3/4\) whereas it follows from Theorem 1.4 that \(\gamma(9, 11)\leq 1/5\).