Hi all,
A notion just occurred to me regarding reducing memory usage. With Strang splitting, all variants of Clawpack currently save the previous time step's q array before advancing by the first half time step of the source term, in case the hyperbolic step has to be rejected. For some problems (I'm thinking of my poroelasticity code here, but the situation is the same for any linear problem and maybe well-behaved nonlinear ones) time steps will never rejected, so I was thinking that it might be worthwhile to add a run time option to not save the previous time step.
I think I'm going to go ahead and add this into my local copy of PyClaw, since it will give me a little more grid size to play with, but is the rest of the PyClaw community interested in this?