Edge Flux Register for Constrained Transport
This Flux Register is useful for solving system like dB/dt + curl E = 0 on a staggered mesh. (Here d is of course partial derivation.) B is a vector on cell faces, and E is a vector on cell edges. In 2D, E has only one component, Ez, and it is on the nodes of a 2d mesh.
At the beginning of a coarse step, reset()
is called. In MFIter for the coarse level advance, CrseAdd
is called with coarse flux (i.e., E). The flux is not scaled. In MFIter for the fine level advance, FineAdd
is called. After the fine level finishes its time steps, Reflux
is called to update the coarse level B on the coarse/fine boundary. The user is also expected to call this version of average_down_faces from AMReX_MultiFabUtil.H to synchronize the coarse level data with the fine level.
\vertbatim template <typename MF, std::enable_if_t<IsFabArray<MF>::value,int>> void average_down_faces (const Array<const MF*,AMREX_SPACEDIM>& fine, const Array<MF*,AMREX_SPACEDIM>& crse, const IntVect& ratio, const Geometry& crse_geom)
Note that both CrseAdd and FineAdd are async for GPU builds. That means it's the user's responsibility to keep the FArrayBox arguments alive or call Gpu::streamSynchronize() when necessary.
Because staggered grids are used, tiling could be very confusing. To avoid confusion, this class assumes that tiling is not enabled for the MFIter loop containing calls to CrseAdd and FineAdd.
If the equation has an extra factor due to the choice of units, the factor can be absorbed into dt. If we have v x B
instead of E, the sign can also been absorbed into dt. Note that whatever the choice of sign is, the dt arguments passed to CrseAdd and FineAdd should have the same sign.
We try to keep the interface simple by not providing overloads that specify the component index. If the user's data does not start with component 0, it can be worked around by creating alias FArrayBox and MultiFab.