On the approximation exponents for subspaces of \(\mathbb{R}^n\) (Q2121976)

Let \(A\) and \(B\) be two subspaces of \(\mathbb R^n\) of respective dimension \(d\) and \(e\) with \(d + e \le n\). Following \textit{W. M. Schmidt} [Ann. Math. (2) 85, 430--472 (1967; Zbl 0152.03602)], the proximity between \(A\) and \(B\) is measured by certain angles, called canonical angles, \(0 \le \theta_1 \le \ldots \le \theta_t \leq \pi/2\) where \(t := \min(d, e)\), and then we set \(\Psi_j(A, B) := \sin(\theta_j) \). When \(B\) is a rational subspace of \(A\) (i.e., \(B\) has a basis of vectors with rational coordinates), consider its height \(H(B) := \text{covol}(B \cap \mathbb Z^n)\). The exponent of approximation \(\mu_n(A\vert e)_j\) is defined as the upper bound (possibly equal to \(+\infty\)) of the set of \(\beta > 0\) such that the inequality \(\Psi_j(A, B) \le H(B)^{-\beta}\) holds for infinitely many rational subspaces \(B\) of dimension \(e\). Consider now the minimal value \(\mathring{\mu}_n(d \vert e)_j\) taken by \(\mu_n(A\vert e)_j\) when \(A\) ranges through the set of subspaces of dimension \(d\) of \(\mathbb R^n\) such that, for all rational subspaces \(B\) of dimension \(e\), one has \(\dim(A \cap B) < j\). Schmidt gave in 1967 explicit lower and upper bounds for \(\mathring{\mu}_n(d\vert e)_j\), which are not always tight. He also computed the exact value of \(\mathring{\mu}_n(d\vert e)_j\) when \(t=1\). Some of Schmidt's upper bounds were improved by Moshchevitin and de Saxé in 2020. The last open case in \(\mathbb R^4\) was the determination of the value of \(\mathring{\mu}_4(2\vert 2)_1\), for which Schmidt's bounds shows that \(3\le \mathring{\mu}_4(2\vert 2)_1 \le 4\). In \(\mathbb R^5\), it was also known that \(4\le \mathring{\mu}_5(3\vert 2)_1 \le 7\). In this paper, the author proves that \(\mathring{\mu}_4(2\vert 2)_1 = 3\) and the refined upper bounds \(\mathring{\mu}_5(3\vert 2)_1 \le 6\) and \(\mathring{\mu}_{2d}(d \vert \ell)_ 1 \le 2d^2 /(2d-\ell)\) for \(d\ge 2\) and \(\ell\in\{1, \ldots, d\}\). He also prove a new lower bound in the general case, from which it follows that \(\mathring{\mu}_n(d \vert d)_d \to 1/d\) as \(n\to +\infty\). Some of these refinements come from a new application of Schmidt's going-up Theorem, which led the author to prove that \(\mathring{\mu}_n(d\vert e)_j\ge \frac{n-\ell}{n-e}\cdot\mathring{\mu}_n(d\vert \ell)_j\) when \(d,e,j,l\ge 1\), \(d+e\le n\), \(1\le j\le \ell\le e \) and \(j\le d\).