A note on squares in arithmetic progressions. II (Q2568671)

An old conjecture of Rudin asserts that there is an absolute constant \(C\) such that any non--constant \(N\)--term arithmetic progression of integers contains at most \(C\sqrt{N}\) perfect squares. In the paper it is shown that this number of squares is smaller than, or at most equal to, \(c_1N^{3/5}(\log N)^{c_2}\), where \(c_1,c_2\) are two positive absolute and computable constants. This reduces the exponent in the previous best known result by \textit{E. Bombieri, A. Granville} and \textit{J.Pintz} [Duke Math. J. 66, 369--385 (1992; Zbl 0771.11034)]. The proof employs many arguments used in this paper but a substantial novelty is that instead of working with algebraic projective curves of genus \(5\) the proof uses elliptic curves.