In some consulting work the problem came up to find the maximum likelihood estimate of the covariance matrix of a multivariate normal distribution in n dimensions, under the constraint that the estimate is bounded below by a known matrix.