Palindromic prefixes and Diophantine approximation (Q2385080)

Uniform simultaneous approximation of a real number and its square has attracted recently the attention of a number of mathematicians, after the work of \textit{D. Roy} [C. R., Math., Acad. Sci. Paris 336, No. 1, 1--6 (2003; Zbl 1038.11042), Proc. Lond. Math. Soc. (3) 88, No. 1, 42--62 (2004; Zbl 1035.11028) and Can. J. Math. 59, 211--224 (2007; Zbl 1115.11036)] who answered a question left open by a paper of \textit{H.~Davenport} and \textit{W. M.~Schmidt} in 1969 [Acta Arith. 15, 393--416 (1969; Zbl 0186.08603)]. Let \(\xi\) be a real number which is not a root of a degree \(2\) polynomial with rational coefficients. Denote by \(\beta_1(\xi)\) the infimum of the set of \(\beta>0\) such that, for any sufficiently large \(B\), there is a triple \({\underline {x}}=(x_0,x_1,x_2)\in\mathbb Z^3\) satisfying \(1\leq x_0\leq B\) and \[ \max\bigl\{ | x_0\xi-x_1|;\;| x_0\xi^2-x_2| \bigr\}\leq B^{-1/\beta}. \] By Dirichlet's box principle, for any \(\xi\in\mathbb R\) not quadratic we have \(\beta_1(\xi)\leq 2\). Also the equality \(\beta_1(\xi)= 2\) holds for almost all \(\xi\in\mathbb R\) (for Lebesgue measure). In the above mentioned paper, Davenport and Schmidt proved the lower bound \(\beta_1(\xi)\geq \gamma\) for any \(\xi\in\mathbb R\) which is not quadratic, where \(\gamma=(1+\sqrt{5})/2=1.618\dots\) is the Golden ratio. 35 years later D.~Roy proved that there exist transcendental numbers \(\xi\) for which \(\beta_1(\xi)= \gamma\), and that the set of values in \((\gamma,2)\) taken by \(\beta_1(\xi)\) is dense. In a previous paper [\textit{S. Fischler}, C. R. Acad. Sci. Paris 339, No. 10, 679--682 (2004; Zbl 1073.11048)] the author introduced variants of the coefficient \(\beta_1\): for any \(\varepsilon\) in the range \(0<\varepsilon<1\), define \(\beta_\varepsilon(\xi)\) as the infimum of \(\beta>0\) as before, but with \(B^{-1/\beta}\) replaced by \[ \min\bigl\{ B^{-1/\beta}; \;| x_0| ^{\varepsilon-1} \bigr\}. \] If the set of such \(\beta\) is empty, then he let \(\beta_\varepsilon(\xi)=+\infty\). The map \(\varepsilon\mapsto \beta_\varepsilon(\xi)\) is non increasing. Hence for \(\varepsilon=0\) let \(\beta_0(\varepsilon)= \sup_{0<\varepsilon<1} \beta_\varepsilon(\xi)= \lim_{\varepsilon\rightarrow 0} \beta_\varepsilon(\xi)\). For almost all \(\xi\in\mathbb R\) we have \(\beta_0(\xi)=+\infty\), and for all \(\xi\) we have \(\beta_0(\xi)\geq \gamma\) by the work of Davenport and Schmidt. Consider the set \[ S=\bigl\{\beta_0(\xi);\;\xi\in\mathbb R;\;\xi \text{ not quadratic} \bigr\}\cap (1,2). \] The author proves that there is an increasing sequence \((\sigma_k)_{k\geq 1}\) of isolated points in \(S\), that converges to the least accumulation point \(\sigma_{\infty}=1.721\dots\) of \(S\), with \(\sigma_1=\gamma\), \(\sigma_2=(2+\sqrt{2})/2=1.707\dots\), \(\sigma_3=(2+\sqrt{10})/3=1.720\dots\) and so on. The proof rests on D.~Roy's palindrome prefix method (see Roy's above mentioned paper as well as the author's paper [\textit{S. Fischler}, J. Comb. Theory, Ser., A, 113, No. 7, 1281--1304 (2006; Zbl 1109.68082)]). Let \(w\) be an infinite word on a finite alphabet. Denote by \((n_i)_{i\geq 1}\) the increasing sequence of lengths of palindromic prefixes of \(w\). Define \(\delta(w)=+\infty\) if this sequence is finite, and \(\delta(w)=\limsup_{i\rightarrow\infty} n_{i+1}/n_i\) otherwise. The main result of the paper under review states that \[ S=\bigl\{ \delta(w); \;w \text{ an infinite word} \bigr\}\cap(1,2). \] The set on the right hand side was already studied by the author (op. cit.), where he proved that its intersection with \((1,\sqrt{3}]\) is the same as the set of these \(\delta(w)\) where \(w\) is a characteristic Sturmian word with \(1\leq \delta(w)\leq 3\). He also uses a result of [\textit{J. Cassaigne}, Theor. Comput. Sci. 218, No. 1, 3--12 (1999; Zbl 0916.68115)] to reach his conclusion.