In 3.2.2, by saying "orthogonal matrix", do we mean that the columns of W are orthonormal? Since W does not look like a square matrix in PCA.