On tests for normality

The problem of deciding whether a sample of a random field was generated by a Gaussian distribution is considered. Based on extensions of large deviation estimates due to M.D. Donsker and S.R.S. Varadhan (1985), a test that is optimal in a generalized Neyman-Pearson sense is proposed. This test turns out to depend on properties of the entropy of Gaussian processes and does not depend on cumulant computations.<
>