Another important feature is that f

Another important feature is that for finding the ground state jwi for each A   B configuration oneemploys some iterative large sparse matrix eigensolver based on sequential applications of bH on someinitial starting vector. To speed up this most time-consuming part of the algorithm, it is highly desirable to have a good prediction for a starting vector, i.e. as close as possible to the ultimate solution.This can be achieved by (approximately) transforming the result of the last step into the shiftedA   B configuration [21] by applying two basis transformations: e.g. A ! A and B ! B for a sweepto the right. The explicit formulae (see [7,21]) can be derived by writing