Setting `which=:LR, nev=1` does not return the generalized eigenvalue with 
the largest real parts, and does not give a warning or error:

n = 10
C = eye(n)
A = zeros(n,n)
A[1] = 100
A[end] = -100
@assert eigs(A, C, nev=1, which=:LR)[1][1] == maximum(eigs(A, C)[1])

Am I expected to set nev greater than the number of eigenvalues I truly 
desire, based on my intuition as a numerical analyst? Or has eigs broken 
its implicit guarantee?

Reply via email to