Some probabilistic remarks on Fermat's last theorem (Q2549720)

Define a measure in the space of all sequences of integers. Let the measure of the set of sequences containing \(n\) have measure \(n^{-\alpha}\). It is easy to see then that for all sequences neglecting a set of sequences of measure \(0\), \(\lim_{k=\infty} a_k/k^{1/(1-\alpha)} = c\). The authors show that for \(\alpha >{2 \over 3}\) with probability one the equation \(a_i+a_j = a_r\) has only a bounded number of solutions but for \(\alpha \leq {2 \over 3}\) it has with probability one infinitely many solutions [cf. \textit{H.Halberstam} and \textit{K.F.Roth}, Sequences. Vol. I. (1966; Zbl 0141.04405)]. Thus speaking very heuristically for \(k>3\) Fermat's last theorem is true with probability one.