We have thus \hat{A}\hat{q}_k=\beta_{k-1}\hat{q}_{k-1}+\alpha_k\hat{q}_k+\beta_k\hat{q}_{k+1}, with \beta_0\hat{q}_0=0 for k=1:n-1 . Remember that the vectors \hat{q}_k are orthornormal and this implies \alpha_k=\hat{q}_k^T\hat{A}\hat{q}_k, and these vectors are called Lanczos vectors.