Lanczos' Algorithm for Sparse Matrices

I just have a quick question about the implementation of lanczos' algorithm. If you are unfamiliar with it, it can be found at .

In the paper, it states that W is a given column n-vector. However, it makes no indication of how it is given and or where you what you would do to calculate it. I wasn't sure exactly where to ask this question, so hopefully somebody here knows.

Thank you very much for your time.