A test for separability in covariance operators of random surfaces (Q2215739)

The paper under review deals with the problem of separability. The authors review the existing tools for validating the assumption of separability for surface data (which is often made in situations where the covariance structure cannot be easily estimated) and, using a setup similar to [\textit{J. A. D. Aston} et al., Ann. Stat. 45, No. 4, 1431--1461 (2017; Zbl 1407.62147)], present an alternative test for the hypothesis of separability in hypersurface data. They derive an explicit expression for the minimal distance between the covariance operator and its approximation by a separable covariance operator (it turns out that this minimum vanishes if and only if the covariance operator is separable) and directly estimate the minimal distance (and not the covariance operator itself) from the available data. As a consequence the calculation of the test statistic does neither use an estimate of the full nonseparable covariance operator nor requires the specification of subspaces used for a projection. The authors also prove the consistency of the proposed estimator of the minimum distance and derive its asymptotic distribution under the null hypothesis \& alternative. These results are used to develop an asymptotic and a bootstrap test for the hypothesis of separability, which are consistent against all alternatives. Next, an investigation of the finite sample properties of the new tests (by means of a simulation study) and a comparison with some existing alternative tests for this problem are conducted.