We present a result under which certain functions of covariance matrices are maximized at multiples of the identity matrix. This is used to show that experimental designs that are optimal under an assumption of independent observations can be minimax, in broad classes of correlation structures.
翻译:暂无翻译