Testing equality between several populations covariance operators (Q1656868)

Given random elements \(X_1,\ldots,X_k\) on a separable Hilbert space, it is natural to ask whether their covariance operators \(\mathbf{\Gamma}_1,\ldots,\mathbf{\Gamma}_k\) are equal. The authors develop a test for comparing these covariance operators based on the Hilbert-Schmidt norm of the difference of the sample covariance operators \[ \widehat{\mathbf{\Gamma}}_i=\frac{1}{n_i}\sum_{j=1}^{n_i} \left(X_{i,j}-\overline{X}_i\right)\otimes\left(X_{i,j}-\overline{X}_i\right)\,, \] for \(1\leq i\leq k\), where \(X_{i,1},\ldots,X_{i,n_i}\) denotes a sample from \(X_i\) with sample mean \(\overline{X}_i\), and \(\otimes\) represents the tensor product. Letting \(n=n_1+\cdots+n_k\), the test statistic is \[ T_{k,n}=n\sum_{j=2}^k\|\widehat{\mathbf{\Gamma}}_j-\widehat{\mathbf{\Gamma}}_1\|^2_\mathcal{F}\,, \] where \(\|\cdot\|_\mathcal{F}\) represents the Hilbert-Schmidt norm. The asymptotic distribution of \(T_{n,k}\) is calculated under the null hypothesis \(H_0:\mathbf{\Gamma}_1=\cdots=\mathbf{\Gamma}_k\) and under certain local alternatives. These distributions are shown to depend on the eigenvalues of an unknown operator, and so the authors propose a bootstrap method to approximate the needed distributions. The paper concludes with two simulation studies and an application to a real data set.