On the mantissa distribution of powers of natural and prime numbers (Q2250816)

The mantissa \(\mathcal {M}(x)\in [1,10)\) of a positive real number is defined by \(\mathcal {M}(x) = x 10^{-\lfloor \log_{10} x \rfloor}\). It is well known that the mantissa distribution \[ F_N(t) = \frac 1N \sum_{n\leq N} {\boldsymbol 1}_{[0,t)}(\mathcal {M}(n^r)) \] of sequences of the form \((n^r)_{n\geq 1}\) (or also \((p_n^r)_{n\geq 1}\), where \(p_n\) denotes the \(n\)-th prime number) has no limiting distribution, that is, the limit \(\lim_{N\to\infty} F_N(t)\) does not exist. However, the logarithmic limiting distribution limit \(\lim_{N\to\infty} F_N^{\log(t)}\), where \[ F_N^{\log(t)} = \frac 1{\log N}\sum_{n\leq N}\frac 1n {\boldsymbol 1}_{[0,t)} (\mathcal {M}(n^r)), \] exists and equals \(\log_{10} t\) (Benford's law). The purpose of this paper is to provide upper bounds of the form \[ \sup_{1\leq t < 10} \left| F_N(t) - \log_{10}t\right| =O \left(\frac 1r \right) \] for the sequence \((n^r)\) and similar bounds. In particular, this means that the distribution \(F_N(t)\) of the sequence \((n^r)\) stabilizes for \(r\) getting larger. The proofs are based on classical analytic number theory methods like the Erdős-Turán inequality and exponential sum estimates.