中文

随机正交矩阵的Fisher信息由高斯矩阵近似

概率论 2025-04-16 v1

摘要

Γn{\Gamma}_nn×nn\times n的Haar不变正交矩阵。设Zn{ Z}_nΓn{\Gamma}_n的左上p×qp\times q子矩阵,Gn{G}_np×qp\times q矩阵,其pqpq个元素为独立标准正态分布,其中ppqq为两个正整数。设L(nZn)\mathcal{L}(\sqrt{n} {Z}_n)L(Gn)\mathcal{L}({G}_n)分别为nZn\sqrt{n} {Z}_nGn{G}_n的联合分布。考虑L(nZn)\mathcal{L}(\sqrt{n} {Z}_n)L(Gn)\mathcal{L}({G}_n)之间的Fisher信息 I(L(nZn)L(Gn))I(\mathcal{L}(\sqrt{n} {Z}_n)|\mathcal{L}(G_n))。本文我们得出结论,当pq=o(n)pq=o(n)时,I(L(nZn)L(Gn))0I(\mathcal{L}(\sqrt{n} {Z}_n)|\mathcal{L}(G_n))\longrightarrow 0,而当c=limnpqn(0,+)c=\lim\limits_{n\to\infty}\frac{pq}{n}\in(0, +\infty)时,则不趋于零。精确地,我们得到当p=o(n)p=o(n)时,I(L(nZn)L(Gn))=p2q(q+1)4n2(1+o(1))I(\mathcal{L}(\sqrt{n} {Z}_n)|\mathcal{L}(G_n))=\frac{p^2q(q+1)}{4n^2}(1+o(1))

关键词

引用

@article{arxiv.2504.10887,
  title  = {Fisher information approximation of random orthogonal matrices by Gaussian matrices},
  author = {Yutong Chen and Yutao Ma and Shuhong Xie and Zhuoya Yao},
  journal= {arXiv preprint arXiv:2504.10887},
  year   = {2025}
}

备注

18 pages