I am using eigenfact to extract the eigenvectors corresponding to the 1000 largest eigenvalues from a symmetric matrix.
V = eigfact(Symmetric(M))[:vectors][,1:1000] V = [:,end:-1:1] #sorts V such that V[:,1] is the eigenvector corresponding to the largest eigenvalue
For some reason, after the 5th eigenvector corresponding to the 5th largest eigenvalue, the precision of the values in the eigenvectors drops dramatically. E.g., 0.2 (what I get in R using
eigen(M, symmetric=TRUE)) vs 0.02 (what I get in Julia). However, the precision in eigenvectors 1 through 5 is excellent.