🚀 go-pugleaf

Dear netters:
 
I have a huge sparse eigensystem Ax=lambda*Bx to solve.  A and
B are 70,000 by 70,000 and they have at most 24 nonzero entries
per row.  But the distribution of nonzero entries doesn't have
any particular pattern in the matrices.  I only need the first
eigenvalue.  What will be an efficient method(in the amount of
storage and computation) to solve this problem?  I only have
700 megabytes of virtual memory to use.  I would very much
like to understand the ideas of the method rather than using 
someone else's code like a black box.  So any thoughtful
suggestions or references of the related literature will be
greatly appreciated.
                                Ching-ju Ashraf