Clean up dual time integration
-
Dual time step should be fully handled in the timestepper (currently requires an externally-registered expression) -
Physical time step should be handled in the timestepper (currently an expression is registered outside) -
Ghost cells should be handled a bit more cleanly. Currently we are hard-coding either 0 or 1. -
norm calculation needs to handle divide-by-zero possibility in ConvergenceMeasure
class. -
higher-order discretization needs to be sorted out -
variable naming -
rotating variables each physical timestep
-
-
sort out BC issues. We should probably only be setting BCs on the new variable (STATE_NP1). The STATE_N variable should come in with BCs set from initialization, and the STATE_NONE (dual-time value) is copied from either STATE_N (on first dual-time step) or STATE_NP1 (on subsequent dual-time steps). -
See if we can get it running on GPU. -
Tolerance should be a specified parameter (with default)