resulting grid depends on number ghost nodes
tommy-engels opened this issue · comments
In the current implementation, the threshold_block routine checks also detail coefficients in the ghost node layer of a block, which is the layer overlapping with neighboring blocks. hence, if a flow feature is near the block interface, but still on the other block, a block can be refined.
this sounded like a secure idea back in the design phase, as it precludes flow features to "arrive" on a block which is not ready for it, i.e. not refined.
an ugly consequence is that the grid is then dependent on ng! Larger overlap possibly refines more blocks. As this is counter-intuitive, it should maybe be removed. In any case, the restriction/prediction cycle is performed on the whole block, including the ghost node layer. It would just be the detail coefficient magnitude check which is altered. This strategy avoids using one-sided interpolation stencils in the block thresholding process.
closed; with most recent changes in master, CDF40 wavelets no longer check ghost node layer for details. hence, grid is independent of number_ghost_nodes