Convergence of continued fraction type algorithms and generators (Q1376015): Difference between revisions

Continued fraction expansions and multidimensional generalizations (including the so-called Jacobi-Perron algorithm) or the ordinary binary expansion give various examples of number-expansion algorithms governed by dynamical systems. This paper exhibits a general setup which generalizes all the above situations. More precisely, let \(D\) be a compact convex set in \({\mathbb{R}}^d\) containing the origin and an interior point. Assume that \(D\) admits a finite or countable partition \({\mathcal D}=(D_k)_k\) which satisfies \(\mu(D_k)>0\) and \(\mu(\partial D_k)=0\), where \(\partial D_k\) denotes the boundary of \(D_k\) and \(\mu\) the normalized Lebesgue measure on \(D\). Now let \(T:D\to D\) be a Borel map such that \(T_k=T_{| _{D_k}}\) is one-to-one from \(D_k\) onto \(D\). For every \(x\in D\) let \((k_m)_{m\geq 1}\) be the symbolic sequence defined by \(k_m=k\) if and only if \(T^{m-1}(x)\in D_k\) and set \(D^{(n)}(x)=\{x\in D\); \(T^mx\in D(k_{m+1})\), \(0\leq m<n\}\). If \(\cap_{n\geq 1}D^{(n)}(x)=\{x\}\) for \(\mu\)-almost \(x\in D\), the partition \({\mathcal D}\) is called a generator. Finally assume that an algorithm \(B\) is given which computes for every point \(x\) in \(D\) a sequence of approximation points \(B^{(n)}(x)\in D\) from the \(n\)-tuple \((k_1,\dots,k_n)\). Next define the sets \(C=\{x\in D\); \((B^{(n)}(x))_n\) converges\(\}\) and \(F=\{x\in D\); \(\lim_n B^{(n)}(x)=x \}\). The authors consider the general case where each \(T_k\) is a linear rational map arising from a \((d+1)\times(d+1)\) matrix \(A(k)\) with \(\det A(k)=\pm 1\). Typically \(T_k(x_1,\dots,d_d)=(y_1,\dots,y_d)\) where \(y_i=(a_{i0}+a_{i1}x_1+\dots +a_{id}x_d)/(a_{00}+a_{i1}x_1+\dots +a_{0d}x_d)\). In that case, \(T^n\) is a linear one-to-one rational map on each \(D^{(n)}(x)=(k_1,\dots,k_n)\). Setting \(B_n(x)= A(k_1)^{-1}\dots A(k_n)^{-1}\), the algorithm \(B\) is defined by \[ B^{(n)}(x)= \bigl((b_n(x))_{1i}/(b_n(x))_{0i},\dots, (b_n(x))_{di}/(b_n(x))_{0i}\bigr). \] The authors introduce the following three statements: (A) \(\mu(C)=1\); (B) \(\mu(F)=1\); (C) \({\mathcal D}\) is a generator. Their main results are: (i) (B) is equivalent to (C); (ii) for \(d=1\), (A), (B), (C) are equivalent; (iii) if (C) is not true or if \(d\geq 2\) then there are examples with \(\mu(F)<1\). Conditions which imply \(\mu(F)=1\) are also given. This basic paper contains useful examples and all proofs are easy to follow. For many other examples, we refer to the book of \textit{F. Schweiger} [Ergodic theory of fibred systems and metric number theory, Oxford (1995; Zbl 0819.11027)].