Hyperbolic summation for functions of the GCD and LCM of several integers (Q6073938)

The article under review deals with asymptotic formulas for sums of the form \[\sum_{n_1 \dotsb n_k \le x} f \Bigl( \left(n_1, \dotsc,n_k \right) \Bigr) \quad \text{and} \quad \sum_{n_1 \dotsb n_k \le x} f \Bigl( \left[n_1, \dotsc,n_k \right] \Bigr),\] where \(\left(n_1, \dotsc,n_k \right)\) and \(\left[n_1, \dotsc,n_k \right]\) are respectively the gcd and the lcm of \(n_1, \dotsc,n_k\), and where the summations run over the hyperbolic region \[\left\{ \left(n_1, \dotsc,n_k \right) \in (\mathbb{Z}_{\ge 1})^k : n_1 \dotsb n_k \le x, \ k \ge 2\right\}.\] For the first sum, it is shown in [\textit{E. Krätzel} et al., Cent. Eur. J. Math. 10, No. 2, 761--774 (2012; Zbl 1285.11120)] that \[\sum_{n_1 \dotsb n_k \le x} f \Bigl( \left(n_1, \dotsc,n_k \right) \Bigr) = \sum_{n \le x} \ \sum_{d^k \delta = n} \; (\mu \star f)(d) \, \tau_k (\delta)\] hence relying this sum to the Dirichlet-Piltz divisor problem. There is no similar identity for the lcm. For multiplicative functions satisfying certain growth conditions, the authors establish precise asymptotic formulas for these sums. Among several examples, the following estimate is proven: let \(r \in \mathbb{R}_{\ge 0}\) be fixed, and \(f\) be a multiplicative function such that there exist constants \(C_1,C_2> 0\) such that \(|f(p) - p^r| \le C_1 p^{r-1/2}\) for all primes, and \(\left| f \left(p^\alpha \right) \right| \le C_2 p^{\alpha r}\) for every prime power \(p^\alpha\) with \(\alpha \ge 2\). Then, for all \(\varepsilon > 0\) small \[\sum_{n_1 \dotsb n_k \le x} f \Bigl( \left[n_1, \dotsc,n_k \right] \Bigr) = x^{r+1} Q_{f,k-1}(\log x) + O \left( x^{r + \theta_k + \varepsilon} \right), \] where \(Q_{f,k-1}\) is a polynomial of degree \(k-1\) and leading coefficient given explicitly be the authors, and \(\theta_k \ge \frac{1}{2}\) is an exponent arising in the Dirichlet-Piltz divisor problem. Numerous examples are given, and a precise formula for the case \(f= \tau\), which does not belong to the afore mentionned class, is also investigated. In this case, the authors prove \[\sum_{n_1 \dotsb n_k \le x} \tau \Bigl( \left[n_1, \dotsc,n_k \right] \Bigr) = x Q_{2k-1}(\log x) + O \left( x^{1 - \frac{2}{2k+1} + \varepsilon} \right), \] where \(Q_{2k-1}\) is a polynomial of degree \(2k-1\) and leading coefficient given explicitly be the authors. The proofs rest on the use of factorizations of some multiple Dirichlet series allowing the authors to decompose the function \(f \Bigl( \left[n_1, \dotsc,n_k \right] \Bigr)\) into a Dirichlet convolution product of some more manageable functions.