Distribution of values of general Euler totient function (Q6063622)

This paper presents an insightful application of the classical Perron formula to asymptotics of certain multiplicative sums with more than one inequality constraints. These are sums of the form \[ \sum_{\substack{n\le X\\ f_i(n) \le Y_i\ (\forall i)} } g(n), \quad (X \to \infty), \] where \(g(n)\) and \(f_i(n)\) are fixed multiplicative functions, and \(Y_i\) are parameters depending on \(X\). The authors' approach contributes a simpler, and avoids certain difficulties in previous, treatment of such sums. Some background. Let \[\Phi(X,Y)= \sum_{\substack{n \le X\\ \varphi(n) \le Y}} 1,\] where \(\varphi(n)\) is the Euler's totient function, one of the most important elementary function in number theory. The study of \(\Phi(X,Y)\) for \(Y=\alpha n\) has a rich history, connecting with probabilistic number theory, dating back to almost a century ago from the extensive line work from I.~J.~Schoenberg, P.~Erdös, S.~D.~Chowla, A.~Walfisz, R.~E.~Dressler, P.~T.~Bateman, H.~G.~Diamond, and G.~Martin, P.~Pollack, and E.~Smith to name a long few. More specifically, in [\textit{H. G. Diamond}, Proc. Sympos. Pure Math. 24, 63--75 (1973; Zbl 0273.10036)] -- the same Proceeding in which H.~L.~Montgomery's impactful Pair Correlation Conjecture [\textit{H. L. Montgomery}, Proc. Sympos. Pure Math. 24, St. Louis Univ. Missouri 1972, 181--193 (1973; Zbl 0268.10023)] also appeared in -- Diamond, in one of his main results, proved that, for \(1 \le Y <X\), one has \[\Phi(X,Y) = X g(Y/X) +E(X,Y),\] where, in particular, \(g(Y/X) = \frac{\zeta(2)\zeta(3)}{\zeta(6)} Y/X + E_2(X, Y), (0< Y/X <1),\) with \(E(X,Y)\) and \(E_2(X, Y)\) explicit error terms. The study of asymptotic for \(\Phi(X,Y)\) also has connection to points on elliptic curves [Algebra Number Theory 8, No. 4, 813-836 (2014; Zbl 1316.11046)]. The main results. In one of the main results of the paper, the authors succeed in recovering Diamond's above mentioned result and show their method applies to a more general class of sums in which \(\Phi(X,Y)\) is a special case. More precisely, let \[\Phi_{k, \beta}(X,Y) = \sum_{\substack{n\le X\\ \frac{\Phi_k(n)}{n^\beta} \le Y}} 1,\] where \[\Phi_k(n) = \left| \left\{ (x_1, x_2, \dots, x_k) \in \left( \mathbb{Z}/n\mathbb{Z} \right)^k\ : \ \mathrm{GCD}(x_1^2 + x_2^2 + \dots + x_k^2, n) = 1 \right\} \right|.\] Thus, \(\Phi_1(n) = \varphi(n)\), and \(\Phi(X,Y) = \Phi_{1, 0}(X,Y)\). The authors obtain asymptotics, with explicit error terms, when (i) \(\beta = k-1\) and \(0 < Y < X\) [Theorem 1.4], (ii) \(\beta \in (k-1, k)\) with \(y\) a specific value depending on \(X\), \(k\), and \(\beta\) [Theorem 1.1], and (iii) \(\beta < k-1\) and \(Y \ge X\) [Theorem 1.5]. Upper bounds are derived for other cases. Results on maximal and minimal orders of \(\Phi_k(n)\) are also obtained [Theorems 1.6, 1.7]. In particular, Theorem 1.4 states the following. Theorem 1.4. Let \(k\ge 1\). For \(0<Y<X\), there exists a positive constant \(r_k\) such that \[ \Phi_{k, k-1}(X, Y) = R_{k, k-1}(1) Y + O \left( Y \exp \left( - \sqrt{ r_k \log Y \log \log Y } \right) \right), \] for all \(X\) sufficiently large (depending on \(k\)), where \(R_{k, k-1}(1)\) is a constant (depending on \(k\)) expressed as a product over primes given explicitly in Eq. (1.8). On substituting \(k=1\), Theorem 1.4 recovers Diamond's result, though with essentially the same order of magnitude in the error term. The proofs of the main results relies on a nice repeated use of Perron's formula, as alluded to earlier. Methods of proof. Perron's formula is a beautiful device used to detect inequalities in number theory. One form of Perron's formula roughly converts the indicator function \(\mathbf{1}_{y>1},\) where \(\mathbf{1}_{y>1} = 1\) if \(y>1\) and \(0\) otherwise, into a truncated integral such as \[\frac{1}{2\pi i} \int_{a-iT}^{a+iT} \frac{y^s}{s} ds + O\left( \frac{y^a}{|\log y|} \frac{1}{T} \right) , (y>1),\] where \(a\) and \(T\) are certain parameters. The authors first keep the condition \(n\le X\) in \(\Phi_{k, \beta}(X,Y)\) outside, replace the condition \(n^\beta Y/ \Phi_k(n) \ge 1\) by Perron's formula above, and then bring the \(n\le X\) sum inside. The authors then apply Perron's formula again to replace the condition \(X/n \ge 1\) which then converts the problem into analytic properties of certain Euler products. This second step is what differs mainly from Diamond's prior treatment, where he used a special integral identity. The usual Cauchy's residue theorem and contour shifting arguments are employed to extract the main term and bound the error terms. The strength of the error terms essentially relies on convexity bound for \(\zeta(s)\). Elaborate analyses of the resulting Euler products, which, in particular, give rise to the various constraints on the parameters in the statements of the Theorems, are carefully carried out. Finally, for an arithmetic function \(f\) and a non-decreasing function \(g\) which is positive for all \(X\) sufficiently large, we say that \(g\) is a maximal order (resp. minimal order) for \(f\) if \[\limsup_{n \to \infty} \frac{f(n)}{g(n)} = 1 \quad \left(\text{resp.}\ \liminf_{n \to \infty} \frac{f(n)}{g(n)} \right).\] For example, the maximal and minimal orders of \(\varphi(n)\) are \(n\) and \(e^{-\gamma}n/\log\log n\), respectively. In the last two results of the paper, the authors utilize partial Euler products to deduce consequences on extremal order of the general Euler totient function \(\Phi_k(n)\). In conclusion, it seems likely that the method of iterating Perron demonstrated here could be applied to analyze other sums with two or more multiplicative inequality constraints, leading to pleasant asymptotics in certain cases.