How many entries of a typical orthogonal matrix can be approximated by independent normals? (Q850976)

The goal of the paper is to find the largest orders such that the left upper block of a random matrix uniformly distributed on \(O(n)\), can be approximated by standard normals. Given two absolutely continuous probability measures \(\mu\) and \(\nu\) on \(\mathbb{R}^m\), the variation distance between \(\mu\) and \(\nu\) is defined as \[ \| \mu-\nu\| =2\cdot\sup_A\,| \mu(A)-\nu(A)| . \] Consider random matrices \(\Gamma_n\) uniformly distributed on \(O(n)\), and let \(Z_n\) be the \(p_n\times q_n\) upper left block of \(\Gamma_n\). Let \(G_n\) be the distribution of \(p_nq_n\) independent standard normals, and let \(\mathcal{L}(\sqrt{n}\,Z_n)\) denote the joint probability distribution of the \(p_nq_n\) entries of \(\sqrt{n}Z_n\). The author proves that if \(p_n=o(\sqrt{n})\) and \(q_n=o(\sqrt{n})\) then \[ \lim_{n\rightarrow\infty}\| \mathcal{L}(\sqrt{n}Z_n)-G_n\| =0, \] and that if \(x,y>0\) are fixed numbers such that \(p_n=[x\sqrt{n}]\), \(q_n=[y\sqrt{n}]\), then \[ \liminf_{n\rightarrow\infty}\| \mathcal{L}(\sqrt{n}Z_n)-G_n\| \geq E\biggl| \exp\biggl(-\frac{x^2y^2}8+\frac{xy}4\,\xi \biggr)-1\biggr| >0, \] where \(\xi\) is a standard normal. For the case of variation distance, the above results show that the largest orders of \(p_n\) and \(q_n\) such that \(Z_n\) can be approximated by standard normals are \(o(\sqrt{n})\). Let \(Y_n\) be \(n\times n\) random matrices with independent standard normal entries, and obtain \(\Gamma_n\) from \(Y_n\) by doing Gram-Schmidt on the columns of \(Y_n\). Then, if \[ \varepsilon_n(m)=\max_{1\leq i\leq n,1\leq j\leq m}\,| \sqrt{n}\gamma_{ij}-y_{ij}| , \] the author proves that \(\Gamma_n\) is Haar invariant on \(O(n)\), that \(\varepsilon_n(m_n)\rightarrow 0\) in probability provided that \(m_n=o(n/\log n)\), and that given \(\alpha>0\), \(\varepsilon_n([n\alpha/\log n])\rightarrow 2\sqrt\alpha\) in probability. The above result shows that \(m_n=o(n/\log n)\) is the largest order such that the entries of the first \(m_n\) columns of \(\Gamma_n\) can be approximated simultaneously by independent standard normals.