Randomness of Mobius coefficents and brownian motion: growth of the Mertens function and the Riemann Hypothesis

Abstract

The validity of the Riemann Hypothesis (RH) on the location of the non-trivial zeros of the Riemann ζ-function is directly related to the growth of the Mertens function M (x) = x k=1 µ(k), where µ(k) is the Möbius coefficient of the integer k: the RH is indeed true if the Mertens function goes asymptotically as M (x) x 1/2+ . We show that this behavior can be established on the basis of a new probabilistic approach based on the global properties of Mertens function. To this aim, we focus the attention on the square-free numbers and we derive a series of probabilistic results concerning the prime number distribution along the series of square-free numbers, the average number of prime divisors, the Erdős-Kac theorem for square-free numbers, etc. These results lead us to the conclusion that the Mertens function is subject to a normal distribution as much as any other random walk, therefore with an asymptotic behaviour given by x 1/2+ . This represents a theoretical advance in the field. We also argue how the Riemann Hypothesis implies the Generalised Riemann Hypothesis for the Dirichlet L-functions. Next we study the local properties of the Mertens function dictated by the Möbius coefficients restricted to the square-free numbers. Motivated by the natural curiosity to see how close to a purely random walk is any sub-sequence extracted by the sequence of the Möbius coefficients for the square-free numbers, we perform a massive statistical analysis on these coefficients, applying to them a series of randomness tests of increasing precision and complexity: together with several frequency tests within a block, the list of our tests include those for the longest run of ones in a block, the binary matrix rank test, the Discrete Fourier Transform test, the non-overlapping template matching test, the entropy test, the cumulative sum test, the random excursion tests, etc. The successful outputs of all these tests (with a level of confidence of 99% that all the sub-sequences analyzed are indeed random) can be seen as impressive "experimental" confirmations of the brownian nature of the restricted Möbius coefficients and the probabilistic normal law distribution of the Mertens function analytically established earlier. In view of the theoretical probabilistic argument and the large battery of statistical tests, we can conclude that while a violation of the RH is strictly speaking not impossible, it is however ridiculously improbable.