A new lower bound for the de Bruijn-Newman constant (Q916716)

Let the function \(H_{\lambda}(x)\), \(\lambda\in {\mathbb{R}}\), be defined by \(H_{\lambda}(x):=\int^{\infty}_{0}e^{\lambda t^ 2}\Phi (t)\cos (xt) dt,\) where \(\Phi (t)=\sum^{\infty}_{n=1}(2n^ 4\pi^ 2e^{9t}-3n^ 2\pi e^{5t})\exp (-n^ 2\pi e^{4t}).\) \textit{C. M. Newman} has shown [Proc. Am. Math. Soc. 61, 245-251 (1977; Zbl 0342.42007)] that there exists a real number \(\Lambda,-\infty <\Lambda \leq\), such that (i) \(H_{\lambda}(x)\) has only real zeroes when \(\lambda\geq \Lambda\), and (ii) \(H_{\lambda}(x)\) has some non-real zeroes when \(\lambda <\Lambda.\) This \(\Lambda\) was baptized the de Bruijn-Newman constant by \textit{G. Csordas}, \textit{T. S. Norfolk} and \textit{R. S. Varga} [Numer. Math. 52, 483-497 (1988; Zbl 0663.65017)]. They proved that \(\Lambda >-50\). The truth of the Riemann hypothesis would imply that \(\Lambda\leq 0\), whereas Newman conjectured that \(\Lambda\geq 0.\) In this note high-precision computations, with a floating-point mantissa of 250 decimal digits, are described which provide strong numerical evidence for the new bound \(\Lambda >-5\). This bound could be proved if one would repeat these computations with an accuracy of about 2600 decimal digits.