5. If the matrix B (the scaled covariance of the population frequencies p) has rank r, then r − 1 of the eigenvalues of that correspond to eigenvectors in F* depend on B. (So if B has full rank, all these eigenvalues depend on B.) If we allow each sample size M(k) → ∞, then then all such eigenvalues also → ∞. (We will call the corresponding eigenvalues the large eigenvalues).
