PCA is optimal in that it returns orthogonal axes in directions of maximum covariance. It can be cast as an optimization problem. It is definitely not optimal

PCA is optimal in that it returns orthogonal axes in directions of maximum covariance. It can be cast as an optimization problem. It is definitely not optimal in a classification sense.