Hypothesis testing on linear structures of high-dimensional covariance matrix (Q2284375): Difference between revisions

Let \(\{{\bar{x}}_1,{\bar{x}}_2,\ldots, {\bar{x}}_n\}\) be an independent and identically distributed random sample from a \(p_n\)-dimensional population \(\bar{x}\) with mean \(\mathbb{E}(\bar{x})=\bar{\mu}\), and covariance matrix \(\mathrm{Cov}(\bar{x})=\Sigma\). Linear structure for a covariance matrix \(\Sigma\) means that this matrix can be represented as a linear combination of the prespecified symmetric \(p_n\times p_n\) matrices \(\{A_1,A_2,\ldots, A_k\}\) with a fixed and finite \(k\), i.e. \[ \Sigma=\theta_1A_1+\theta_2A_2+\ldots+\theta_kA_k, \] where \(\theta_1, \theta_2,\ldots, \theta_k\) are unknown parameters. The authors of the paper develop tests for the linear structure of covariance matrices. The main results of the paper are obtained by supposing that the sequence \(p_n/(n-1)\) converges to a positive number and that \(\bar{x}=\bar{\mu}+\Sigma^{1/2}\bar{w}\), where the vector \(\bar{w}\) consists of independent and identically distributed components with zero mean, unit variance and finite moment of order four.