A block-Lanczos method for large continuation problems

The computation of solution paths of large-scale continuation problems can be quite challenging because a large amount of computations have to be carried out in an interactive computing environment. The computations involve the solution of a sequence of large nonlinear problems, the detection of turning points and bifurcation points, as well as branch switching at bifurcation points. These tasks can be accomplished by computing the solution of a sequence of large linear systems of equations and by determining a few eigenvalues close to the origin, and associated eigenvectors, of the matrices of these systems. We describe an iterative method that simultaneously solves a linear system of equations and computes a few eigenpairs associated with eigenvalues of small magnitude of the matrix. The computation of the eigenvectors has the effect of preconditioning the linear system, and numerical examples show that the simultaneous computation of the solution and eigenpairs can be faster than only computing the solution. Our iterative method is based on the block-Lanczos algorithm and is applicable to continuation problems with symmetric Jacobian matrices.