No label defined (Q4252249)

Let \(f : {\mathbb N} \to {\mathbb C} \) be an additive function; the truncated functions \( f_y \), defined by \[ f_y(n)= \sum_{p \leq y, p^v\|n} f(p^v), \] ``behave like'' independent random variables \(Z_y= \sum_{p \leq y} \xi_p, \text{where } P(\xi_p= f(p^v))= (1- {1 \over p}) \cdot p^{-v}\). More precisely, \(K(x,y)= \sup_{f \text{ additive}} d_{TV} (f_y, Z_y) \to 0\), as \(u= {\log x \over \log y} \to \infty\), where \(d_{TV} (f_y, Z_y)= \sup_{A \subset {\mathbb R}} |\nu_x (f_y \in A)- P(Z_y \in A)|\). Improving on results of Kubilius (1956), Barban and A. I. Vinogradov (1964) and Arratia and Stark (1997), the author gives (Theorem 1) an optimal upper estimate \[ K(x,y) \ll_\varepsilon \rho(u) \cdot 2^{(1+\varepsilon)u}+ x^{-1+\varepsilon}, \] where \( \rho \) is Dickmann's function, and an asymptotic formula \[ K(x,y)= H(u) \cdot \left\{1+ {\mathcal O} \left({\log(u+1)+ 2^{-u}\log\log y \over \log y} \right) \right\}+ {\mathcal O} \left({\rho^2 ({1 \over 2}u) \over Y_\varepsilon} \right) \] uniformly in \((H_\varepsilon)\) \(x \geq 3\), \(\exp((\log\log x)^{{5 \over 3}+\varepsilon}) \leq y \leq x\). The function \( H(\cdot)\) is (according to Arratia and Stark) \[ H(u)= \lim_{y \to \infty} K(y^u, y)= {1 \over 2} \int^\infty_{-\infty} |\omega(v)-e^{-\gamma} |\cdot \rho(u-v)dv+ {1 \over 2} \rho(u), \] with Buchstab's function \( \omega \). In particular, in \( x \geq 3, \exp\left((\log x)^{{2 \over 5}+\varepsilon} \right) \leq y \leq x\), \[ K(x,y)= H(u) \cdot \left\{1+ {\mathcal O} \left({\log\log y \over \log y} \right) \right\}. \] These results imply an improved estimate of the Lévy distance of the distribution functions \(F_x(z)= \nu_x (f(n) \leq A_x+ z B_x)\), \(G_x(z)= P(Z_x \leq A_x+ zB_x)\), and an asymptotic formula [valid in \(x \geq 2\), \(y \geq 2\), \(1 \leq z \leq \exp (e^{(\log y)^{{3 \over 5}- \varepsilon}})]\) for \(\Theta (x,y,z)= \#\{n \leq x, n_y > z\}\), where \( n_y \) is the largest divisor of \(n\) with largest prime divisor \(\leq y \). For the proof of this main result, the author gives estimates (or asymptotic formulae) for expressions like \( \omega(u)- e^{-\gamma}\), \(r'(u)= {d \over du} (- {\rho'(u) \over \rho(u)})\), \(\rho({1 \over 2} u)\) in terms of \( \rho(u) \) and \( \xi(u)\) (defined by \(e^{\xi(u)}= 1+ u \xi(u))\), \(H(u)\), \({x^s \zeta(s) \over \prod_{p \leq y}(1-p^s)^{-1}}\), and he gives (Theorem 2) an asymptotic formula for \[ \Phi(x,y)= \#\{n \leq x; \;p|n \Rightarrow p> y\}, \] uniformly valid in \( x \geq y \geq 2\) [the critical range is \(\exp \{ (\log x)^{{2 \over 5}+\varepsilon}\}\leq y \leq \exp \{ (\log x)^{{5 \over 8}+ \varepsilon}\}]\). In particular, uniformly in \( (H_\varepsilon) \), \[ \Phi(x,y) = {x \over \prod_{p \leq y} \left( 1 - {1 \over p}\right)^{-1}} + {\mathcal O} \left( {x R(u) \over \log y} + {x \rho(u) \over Y_\varepsilon} \right), \] and \[ \Phi(x,y) = {e^\gamma (x \omega(u) - y) \over \prod_{p \leq y} \left( 1 - {1 \over p}\right)^{-1}} + {\mathcal O} \left( {x R(u) \log (u+1) \over (\log y)^2} + {x \rho(u) \over Y_\varepsilon} \right), \] where \[ R(u) = \rho(u) \cdot \exp \left\{ {-\pi^2 u \over 2 \xi^2(u)} + {\mathcal O} \left( {u \over \xi^3(u)} \right) \right\}, \] and \( Y_\varepsilon = \exp \{(\log y)^{{3 \over 2} - \varepsilon}\}. \) Moreover, there is an asymptotic formula \[ \sum_{1 < n \leq x} \log P^{-}(n) = e^{-\gamma} \cdot x \cdot \log\log x + A \cdot x + {\mathcal O} \left( x \cdot \exp \left\{ -(\log x)^{{3 \over 8} + \varepsilon}\right\}\right), \] for the least prime factor \( P^{-}(n) \) of \(n\). For the entire collection see [Zbl 0911.00018].