Behaviour of log \(\eta\) (\(\tau\) ) at a rational point (Q1087575)

Let \(\eta(\tau)\) be the Dedekind \(\eta\)-function; with \(Im(\tau)>0\). Fix \(\log\eta(\tau) = \pi i\tau /12-\sum^{\infty}_{n=1} q^{2n}(1- q^{2n})/n\); with \(q=e^{\pi i\tau}\). Write \(\tau =b/a+iy\), where \(a,b\in {\mathbb{Z}}\), \((a,b)=1\). \textit{G. H. Hardy}, using the circle method [Q. J. Math. 34, 76-86 (1902), Collected papers, Vol. IV, pp. 351-361 (Oxford 1969; Zbl 0185.289)], showed that \[ \log \eta (b/a+iy)=-\pi /12a^ 2y-\log (ay)-i(s(b,a)-b/12a)+o(1)\text{ as } y\to 0^+. \] Here \(s(b,a)\) is the classical Dedekind sum. This paper uses the Euler-Maclaurin summation formula, and not the transformation formula for \(\eta\), to show that the \(o(1)\) may be replaced by a term \(\ll y^ n\), for any integer \(n>0\). (The depenence of the implied constant on a and b is not made clear.) Using the \(\eta\) transformation formula, the authors improve this to \(\ll e^{-2\pi /a^ 2y}\), where \(2\pi\) is best possible.