On transfer inequalities in Diophantine approximation. II (Q971906)

This paper is a continuation of the previous work by the second author [Analytic number theory. Essays in honor of Klaus Roth on the occasion of his 80th birthday. Cambridge: Cambridge University Press, 306--314 (2009; Zbl 1163.11053)]. The classical Khintchine's transference principle relates the sharpness of simultaneous rational approximation to an \(n\)-tuple of real numbers \(\theta_1,\dots,\theta_n\) and the measure of linear independence of \(1,\theta_1,\dots,\theta_n\) over \(\mathbb Q\). In the paper under review, the authors refine Khintchine's theorem. To state the authors' main result, some notation is needed. For a point \(\Theta := (\theta_1,\dots,\theta_n) \in \mathbb R^n\), define \(\omega_0(\Theta)\) and \(\omega_{n-1}(\Theta)\) respectively to be the supremum, possibly infinite, of \(\omega \in \mathbb R\) for which there exist infinitely many \((x_0,\dots,x_n) \in \mathbb R^{n+1}\) with \[ \max_{1 \leq i \leq n} |x_0\theta_i-x_i| \leq \left( \max_{0 \leq i \leq n} |x_i| \right)^{-\omega} \text{ or } |x_0+x_1\theta_1+\dots+x_n\theta_n| \leq \left( \max_{0 \leq i \leq n} |x_i| \right)^{-\omega}. \] Then Khintchine's transference principle states that the inequalities \[ \frac{\omega_{n-1}(\Theta)}{(n-1)\omega_{n-1}(\Theta)+n} \leq \omega_0(\Theta) \leq \frac{\omega_{n-1}(\Theta)-n+1}{n} \] hold for any \(\Theta = (\theta_1,\dots,\theta_n) \in \mathbb R^n\) with \(1,\theta_1,\dots,\theta_n\) linearly independent over \(\mathbb Q\). It is also known from the work of Jarník that these inequalities are best possible. The purpose of the paper under review is to show that nevertheless these inequalities can be refined by an introduction of additional uniform exponents \(\hat{\omega}_0(\Theta)\) and \(\hat{\omega}_{n-1}(\Theta)\). Specifically, let \(\hat{\omega}_0(\Theta)\) and \(\hat{\omega}_{n-1}(\Theta)\) respectively be the supremum of \(\omega \in \mathbb R\) such that for all sufficiently large \(X \in \mathbb R\), there exists \((x_0,\dots,x_n) \in \mathbb Z^{n+1} \setminus \{\mathbf 0\}\) with \[ \max_{0 \leq i \leq n} |x_i| \leq X, \] satisfying the inequality \[ \max_{1 \leq i \leq n} |x_0\theta_i-x_i| \leq X^{-\omega} \text{ or } |x_0+x_1\theta_1+\dots+x_n\theta_n| \leq X^{-\omega}. \] The authors then prove that when \(n \geq 2\), the inequalities \[ \frac{(\hat{\omega}_{n-1}(\Theta)-1)\omega_{n-1}(\Theta)}{((n-2)\hat{\omega}_{n-1}(\Theta)+1)\omega_{n-1}(\Theta)+(n-1)\hat{\omega}_{n-1}(\Theta)} \leq \omega_0(\Theta) \] and \[ \omega_0(\Theta) \leq \frac{(1-\hat{\omega}_0(\Theta))\omega_{n-1}(\Theta)-n+2-\hat{\omega}_0(\Theta)}{n-1} \] hold for any \(\Theta = (\theta_1,\dots,\theta_n) \in \mathbb R^n\) with \(1,\theta_1,\dots,\theta_n\) linearly independent over \(\mathbb Q\). These inequalities constitute an improvement of Khintchine's theorem, since \[ \hat{\omega}_{n-1}(\Theta) \geq n,\;\hat{\omega}_0(\Theta) \geq \frac{1}{n}. \] Furthermore, the authors also introduce intermediate exponents \(\omega_d(\Theta)\) for \(d=1,\dots,n-2\), measuring the accuracy of approximating a point \(\Theta \in \mathbb R^n\) by rational linear subvarieties of dimension \(d\). They also discuss transference inequalities linking \(\omega_d(\Theta)\) and \(\omega_{d+1}(\Theta)\), the composition of which gives Khintchine's theorem.