Effcient Solution of Evolution Models for Virus Populations

The computation of the quasispecies in Eigen's quasispecies model requires the solution of a very large scale eigenvalue problem. Since the problem dimension is of an exponential growing nature the well known methods for dealing with such a problem run out of resources already far away from practically relevant cases. We propose the use of an implicit matrix vector product using the special problem structure as building block for eigenvalue solvers to partially overcome the exponential growth, which let us reach unexpected large problem sizes. As we will show our implicit matrix vector product is a prime example for an algorithm perfectly matching the requirements of GPU computing since it has low space and high parallel computation requirements. Therefore we will also present an GPU implementation delivering a speedup factor of about 100 compared to a standard implementation.