The image of Carmichael's \(\lambda\)-function (Q482304)

Let \(\lambda(m)\) denote Carmichael's function, defined as the smallest natural number \(n\) such that \[ a^{n} \equiv 1 \mod m \quad \forall (a,m)=1 . \] Thus \(\lambda(m)\) is closely related to the Euler totient function \(\varphi\), but unlike \(\varphi\) it is not multiplicative, instead satisfying \[ \lambda(p_1^{a_1} \cdots p_{w}^{a_w}) = \text{lcm}(\lambda(p_1^{a_1}), \ldots, \lambda(p_w^{a_w})) \] for any distinct primes \(p_i\). One can also note that, since primitive roots always exist mod \(p\), we have \(\lambda(p)=p-1\) for all primes \(p\). Now let \[ V_{\lambda}(x) := \#\{n \leq x: n = \lambda(m) \; \text{for some} \; m\} . \] In this paper the authors prove that \[ V_{\lambda}(x) \geq \frac{x}{(\log x)^{\eta + o(1)}} \;\;\; \text{as} \; x \rightarrow \infty , \] where \(\eta = 1 - \frac{(1+\log\log 2)}{\log 2} \approx 0.086\) is the multiplication table constant, which improves on various earlier lower bounds. It was previously shown by the second and third authors [Acta Arith. 162, No. 3, 289--308 (2014; Zbl 1292.11109)] that \(V_{\lambda}(x) \leq \frac{x}{(\log x)^{\eta + o(1)}}\), so the behaviour of \(V_{\lambda}(x)\) is now determined up to the ``little oh'' term in the exponent of log. The proof works by counting some special values of \(n\) in the image of the \(\lambda\) function, using sieve methods, basic results on the distribution of integers with a given quantity of prime factors, and ultimately a Cauchy-Schwarz argument. In particular, these \(n\) are squarefree and have about \((\log\log x)/\log 2\) prime factors. More precisely, let \(k \geq 2\) be a fixed integer, and for each squarefree \(n\) let \(r(n)=r(n,k)\) denote the number of representations of \(n\) in the form \[ n = \prod_{i=0}^{k-1}a_i \cdot \prod_{j=1}^{2^{k}-1}b_j , \] where (roughly speaking) the \(a_i \) are bigger than 1 and have all their prime factors bigger than \(y\); the \(b_j\) each have precisely \(l\) prime factors, all of which are \(\leq y\); and where all of the numbers \(a_i B_i + 1\) for \(0 \leq i \leq k-1\) are prime, where \(B_i := \prod_{j : \lfloor j/2^{i} \rfloor \; \text{odd}} b_j\). Here \(y\) and \(l\) are certain parameters depending on \(x\) and \(k\) (where, in particular, \(l\) is chosen to ensure that \(n\) ends up having about \((\log\log x)/\log 2\) prime factors). If \(n\) is squarefree and \(r(n) > 0\) then all of the primes \(a_i B_i + 1\) must be distinct, and we have \[ n = \text{lcm}(a_0 B_0, \ldots, a_{k-1}B_{k-1}) = \lambda(\prod_{i=0}^{k-1} (a_i B_i + 1)) . \] Thus one can lower bound \(V_{\lambda}(x)\) by noting that, by Cauchy-Schwarz, \[ V_{\lambda}(x) \geq \frac{(\sum_{n \sim x} \mu^{2}(n) r(n))^2}{\sum_{n \sim x} \mu^{2}(n) r(n)^2} . \] A lower bound for the first sum, and an upper bound for the second one, can be obtained using the sieve (initially fixing the \(b_j\), then counting possible choices of the \(a_i\), and then counting possible choices of the \(b_j\), essentially), and the authors' theorem follows on finally letting \(k \rightarrow \infty\). As the authors explain, a version of this argument with the fixed choice \(k=2\) was used earlier by Luca and Pomerance [loc. cit.] to give a weaker lower bound for \(V_{\lambda}(x)\). \textit{Further remarks:} The paper is very interesting and the arguments seem clear and clever. There seem to be a small number of typos that the reader might like to keep in mind: following display (4.2), the authors presumably mean \(2^k - 1 + k\) tuples rather than \(2^{k-1}+k\) tuples; in the display preceding (4.7), it should be \(\mu^{2}(2b')\) rather than \(\mu^{2}(b')\), which would slightly complicate the subsequent calculation; the third condition in Lemma 5.1 doesn't seem to be expressed quite correctly (as it stands it can never be satisfied because of the case \(i=j\)); and following display (6.2), the authors presumably mean \((a_0',\ldots,a_{k-1}')\) rather than \((a_0',\ldots,a_{k-1})\).