Euclidean algorithms are Gaussian over imaginary quadratic fields (Q6879285)

It is well known that the Euclidean algorithm can be recorded via a continued fraction (CF). That is \( {a}/{b}:=[a_1,a_2, \dots, a_{\ell}]. \) For a rational \(x \in [0,1]\cap \mathbb{Q}\), we let \(\ell(x):=\ell\) and call it the length of the continued fraction of \(x\). Let \(\mathcal{O} \subset \mathbb{C}\) be a discrete ring for which the field of fractions is Euclidean. It is one of the imaginary quadratic fields \(K_d:=\mathbb{Q}(\sqrt{-d})\) for \(d=1,2,3,7,11\). It is found a strict fundamental domain \( I'_d \subset \{z \in \mathbb{C} : |z|\leqslant 1\}\) of the translation action of \(\mathcal{O}_d\) on \(\mathbb{C}\) for such \(K_d\) and its ring of integers \(\mathcal{O}_d\). An analogue of the Gauss map is defined. It is called the Hurwitz CF map or the nearest integer complex CF map. A dynamical framework for the statistical study of \(K\)-rational trajectories based on the transfer operator methods is presented.\N\NThe following theorem is one of the main results of the article.\N\N\textbf{Theorem.} Let \(c : \mathcal{O} \to \mathbb{R}_{\geqslant 0}\) be bounded and not cohomologous to zero. The distribution of the total cost \(C\) on\N\[\N\Omega_N := \left\{\frac{a}{b} \in I : |b|^2 < N \right\}\N\]\Nis asymptotically Gaussian, that is, there exist real numbers \(\mu(c), \delta(c)>0\) such that for any \(u \in \mathbb{R}\),\N\[\N\mathbb{P}_N \left[\frac{C-\mu(c) \log N}{\delta(c)\sqrt{\log N}}\leqslant u\right]=\frac{1}{\sqrt{2\pi}} \int_{-\infty}^u e^{-\frac{t^2}{2}} dt +O\left(\frac{1}{\sqrt{\log N}} \right)\N\]\Nas \(N \to \infty\), where \(\mathbb{P}_N\) denotes the uniform probability measure on \(\Omega_N\).The expectation and variance satisfy\N\[\N\mathbb{E}_N [C|\Omega_N]=\mu(c) \log N+\mu_1(c)+ O(N^{-\gamma}),\N\]\N\[\N\mathbb{V}_N [C|\Omega_N]= \delta(c) \log N+\delta_1(c)+O(N^{-\gamma}),\N\]\Nfor some \( \mu_1(c), \delta_1(c)\), and \(\gamma>0\).