The Laguerre inequalities with applications to a problem associated with the Riemann hypothesis (Q1187058): Difference between revisions

The Riemann \(\xi\)-function has the integral representation \(\xi(x/2)=8\int^\infty_0\Phi(t)\cos xt \.dt\). It was shown by \textit{G. Csordas}, \textit{T. Norfolk} and \textit{R. Varga} [Numer. Math. 52, 483--497 (1988; Zbl 0663.65017)] that, for each real \(\lambda\), the function \(H_\lambda (x)=\int^\infty_0e^{\lambda t^2}\Phi(t)\cos xt \,dt\) is an even real entire function of order 1 and of maximal type. Since \(H_0(x)=\xi(x/2)/8\), the Riemann Hypothesis is equivalent to the statement that all the zeros of \(H_0(z)\) are real. From the works of \textit{N. G. de Bruijn} [Duke Math. J. 17, 197--226 (1950; Zbl 0038.23302)] and \textit{C. M. Newman} [Proc. Am. Math. Soc. 61, 245--251 (1977; Zbl 0342.42007)] there is the de Bruijn-Newman constant \(\Lambda\leq \tfrac12\) such that \(H_\lambda(x)\) has only real zeros if and only if \(\lambda\ge\Lambda\), and that \(H_\lambda(x)\) has some complex zeros if and only if \(\lambda<\Lambda\). Thus, if the Riemann Hypothesis is true, then \(\Lambda\le 0\). Csordas, Norfolk and Varga first constructed the lower bound \(-50<\Lambda\), which was subsequently improved to \(-5<\Lambda\), and to \(-0.385<\Lambda\) by \textit{H. J. J. te Riele} [Numer. Math. 58, 661--667 (1991; Zbl 0715.11074)], and by \textit{R. Varga}, \textit{T. Norfolk} and \textit{A. Ruttan} [``A lower bound for the de Bruijn-Newman constant \(\Lambda\). II'' in: Conf. Proc. US-USSR Conf. Approximation Theory, Tampa1990, Springer Ser. Comput. Math. 19, 403--418 (1992; Zbl 0787.30016)], respectively. Here the authors give a new method for finding lower bounds for \(\Lambda\), and to apply it to give the new bound \(- 0.0991<\Lambda\). The new method involves functions in the ``Laguerre-Pólya class'' which has the property that \(H_ \lambda\) is in the class if and only if \(\lambda\ge\Lambda\). The authors then introduce the Laguerre differences for a function \(f\), the first difference being given by \(L_1(f(x))=(f'(x))^2-f(x)f''(x)\), and Csordas and Varga established the Laguerre inequality that \(L_1(f(x))\ge 0\) for suitable \(f(x)\). Thus, if \(L_1(H_\lambda(x))<0\) for some real \(\lambda\) and \(x\), then \(\lambda<\Lambda\). Using the information that the two consecutive zeros \(\rho_ n\), \(\rho_{n+1}\) of the Riemann \(\zeta\)-function are particularly close together when \(n=212\), the authors use an iterative scheme which settles on \(\lambda=\hat\lambda=-0.0991\), and set \(x\) to be about \(-830\). However, the rigorous proof that \(L_1(H_ \lambda(x))<0\) is rather involved, mainly because of the high precision required for the evaluation of functions. Since the method is based on real arithmetic, it has decided computational advantage over previous methods. The authors also remarked that there is every reason to believe that the application of the method will give even better results by using smaller differences \(|\rho_n-\rho_{n+1}|\), and a table of such small differences is given for \(n\le 2 000 000\) from well known calculations of \(\rho_ n\). Such improvement will involve the evaluation of functions in much higher precision than the 210 floating point arithmetic being used here. The paper gives a very well motivated account of the method, together with explanation of mathematical requirement in order to secure the sufficient accuracy for function evaluations.