jdqr accepts an approximate Schur vector q with associated eigenvalue lambda if the 2-norm of the residual r is less than Options.Tol. Then we have that
How accurate the approximations of the eigenvectors and the invariant subspaces are, depends on the conditioning of these quantities.
The residual that jdqr computes is given by r = (I-Q*Q')*(A*q-theta*B*q).
