Optimality of two inequalities for exponents of Diophantine approximation (Q2112779): Difference between revisions

We use exponents of approximation to show how well we can approximate real \(n\)-tuples, \(\underline{\xi} = \left( \xi_{1}, \ldots, \xi_{n} \right) \in {\mathbb R}^{n}\), by \(n\)-tuples of rational numbers. The classical exponents of approximation are defined as follows. Let \(\omega\) and \(\widehat{\omega}\) be the suprema of real numbers, \(u\), such that the system \[ 1 \leq x \leq X, \quad \max_{1 \leq i \leq n} \left| x\xi_{i}-y_{i} \right| \leq X^{-u} \] has a solution \(\left( x, y_{1}, \ldots, y_{n} \right) \in {\mathbb Z}^{n+1}\) for arbitrarily large \(X\) and for all large \(X\), respectively. Analogously, we let \(\omega^{*}\) and \(\widehat{\omega}^{*}\) be the suprema of \(v\) such that \[ 1 \leq \max_{1 \leq i \leq n} \left| a_{i} \right| \leq X, \quad \left| a_{0}+a_{1}\xi_{1}+\cdots +a_{n}\xi_{n} \right| \leq X^{-v} \] has a solution \(\left( a_{0}, a_{1}, \ldots, a_{n} \right) \in {\mathbb Z}^{n+1}\) for arbitrarily large \(X\) and for all large \(X\), respectively. For \(n \geq 2\), \textit{Y. Bugeaud} and \textit{M. Laurent} [Math. Z. 265, No. 2, 249--262 (2010; Zbl 1234.11086)] proved that every \(\underline{\xi} \in {\mathbb R}^{n}\) satisfies both \[ \omega \geq \frac{\left( \widehat{\omega}^{*}-1 \right) \omega^{*}}{\left( (n-2)\widehat{\omega}^{*}+1 \right)\omega^{*}+(n-1) \widehat{\omega}^{*}} \] and \[ \omega^{*} \geq \frac{(n-1)\omega+\widehat{\omega}+n-2}{1-\widehat{\omega}}. \] For \(n=2\), both of these estimates are sharp. When \(\omega^{*} \geq n\) and \(\widehat{\omega}^{*}=n\), the first of these inequalities is sharp. Similarly, for \(\omega \geq 1/n\) and \(\widehat{\omega}=1/n\), the second inequality is sharp. Otherwise, the optimality of these inequalities is not well understood for \(n \geq 3\). The author investigates this problem in this paper. He gives a complete description of when these two inequalities are actually equalities. Theorem~2.1 contains the result for the first inequality. For any \(w^{*} \in [n, \infty]\), \(w\) in a closed interval depending only on \(n\) and \(w^{*}\) (precisely defined by the author) and \(\widehat{w}^{*}=w^{*}(w+1)/\left( w^{*}-(n-2)ww^{*}-(n-1)w \right)\), there are \(\underline{\xi} \in {\mathbb R}^{n}\) such that the first inequality holds. In addition, for each such triple \(\left( w^{*}, \widehat{w}^{*}, w \right)\), the author obtains lower bounds for the Hausdorff dimension (\(\geq n-2\)) and the packing dimension (\(\geq n-2+1/n\)) of the set of such \(\underline{\xi}\). Theorem~2.2 provides the analogous results for the optimality of the second inequality above. The proof is based on the parametric geometry of numbers, developed by \textit{W. M. Schmidt} and \textit{L. Summerer} [Monatsh. Math. 169, No. 1, 51--104 (2013; Zbl 1264.11056)], with the variational principle in the parametric geometry of numbers playing an important role.