The asymptotic distribution of Frobenius numbers (Q977169)

Let \(d \geq 2\) be an integer, and define \[ \hat{\mathbb Z}^d_{\geq 2} := \{ {\mathbf a} = (a_1,\dots,a_d) \in {\mathbb Z}^d : \gcd(a_1,\dots,a_d) = 1,\;a_i \geq 2\;\forall\;1 \leq i \leq d \}. \] It is well known that for each fixed \({\mathbf a} \in \hat{\mathbb Z}^d_{\geq 2}\), any sufficiently large positive integer \(N\) can be represented as \[ N = {\mathbf m} \cdot {\mathbf a}, \] where \({\mathbf m} \in {\mathbb Z}^d_{\geq 0}\). In fact, the Frobenius number of \({\mathbf a} \in \hat{\mathbb Z}^d_{\geq 2}\), denoted by \(F({\mathbf a})\), is defined to be the largest positive integer that has no such representation, i.e., \[ F({\mathbf a}) := \max {\mathbb Z} \setminus \{ {\mathbf m} \cdot {\mathbf a} > 0 : {\mathbf m} \in {\mathbb Z}^d_{\geq 0} \}. \] No closed form explicit formulas for the Frobenius number are known when \(d > 2\), and in the recent years a significant amount of effort has been invested by different authors into the investigation of its average behavior. In the current paper, the author extends the recent studies in case \(d=3\) by \textit{V. I. Arnold} [Funct. Anal. Appl. 33, No. 4, 292--293 (1999); translation from Funkts. Anal. Prilozh. 33, No. 4, 65--66 (1999; Zbl 1042.11064)], \textit{J. Bourgain} and \textit{Y. Sinai} [Russ. Math. Surv. 62, No. 4, 713--725 (2007); translation from Usp. Mat. Nauk 62, No. 4, 77--90 (2007; Zbl 1151.11046)], and \textit{V. Shchur, Y. Sinai}, and \textit{A. Ustinov} [J. Number Theory 129, No. 11, 2778--2789 (2009; Zbl 1229.11050)]. He shows that if \({\mathbf a}\) is taken to be random in an expanding \(d\)-dimensional domain, then \(F({\mathbf a})\) has a limit distribution, given by the probability distribution for the covering radius of a certain simplex with respect to a \((d-1)\)-dimensional random lattice. More specifically, let \(d \geq 3\) and let \(\mathcal D \subset \mathbb R^d_{\geq 0}\) be any bounded set with boundary of Lebesgue measure zero. Then the author proves that for any \(R \geq 0\), \[ \lim_{T \to \infty} \frac{1}{T^d} \# \left\{ {\mathbf a} \in \hat{\mathbb Z}^d_{\geq 2} \cap T\mathcal D : \frac{F({\mathbf a})}{(a_1 \dots a_d)^{1/(d-1)}} > R \right\} = \frac{\text{Vol}(\mathcal D)}{\zeta(d)} \Psi_d(R), \] where \(\Psi_d : \mathbb R_{\geq 0} \to \mathbb R_{\geq 0}\) is a non-increasing function with \(\Psi_d(0)=1\). In fact, the author shows that \(\Psi_d(R)\) is equal to the value of the unique \(\text{SL}(d-1,\mathbb R)\)-right invariant probability measure of the set \[ \left\{ A \in \text{SL}(d-1,{\mathbb Z}) \setminus \text{SL}(d-1,\mathbb R) : \rho(A) > R \right\}, \] where \(\rho(A)\) is the covering radius of the simplex \[ \Delta = \left\{ \mathbf{x} \in \mathbb R^{d-1}_{\geq 0} : \sum_{i=1}^{d-1} \leq 1 \right\} \] with respect to the lattice \({\mathbb Z}^{d-1} A\), i.e., \[ \rho(A) = \inf \left\{ \rho \in \mathbb R_{> 0} : {\mathbb Z}^{d-1} A + \rho\Delta = \mathbb R^{d-1} \right\}. \] As the author points out, the expression for \(\Psi_d(R)\) connects naturally with the optimal lower bound \[ F({\mathbf a}) \geq (a_1 \dots a_d)^{1/(d-1)} \inf \left\{ \rho(A) : A \in \text{SL}(d-1,{\mathbb Z}) \setminus \text{SL}(d-1,\mathbb R) \right\}, \] obtained by \textit{I. Aliev} and \textit{P. Gruber} [J. Number Theory 123, No. 1, 71--79 (2007; Zbl 1114.11025)]. The author's method rests on a novel interpretation of \(F({\mathbf a})\) in terms of the dynamics of a certain flow on the space of lattices \(\text{SL}(d-1,{\mathbb Z}) \setminus \text{SL}(d-1,\mathbb R)\) and an equidistribution theorem for a multidimensional Farey sequence on closed horospheres. In the appendix, the author points out how this equidistribution result provides a new perspective on the well-known work of \textit{W. M. Schmidt} on distribution of sublattices of \({\mathbb Z}^d\) [Monatsh. Math. 125, No. 1, 37--81 (1998; Zbl 0913.11028)].