Fix field reflection across lower z #364
Conversation
germasch
left a comment
germasch
left a comment
There was a problem hiding this comment.
Well, staggered grids suck...
For some historical context, PSC used to always use three ghost points on either side, which I think was meant such that it covered [-2, -1, 0, ..., m-1, m, m+1] and [-1.5, -.5, .5, .., m-.5, m+.5, m+1.5] depending on staggering (so it would actually 2 ghost on the left and 3 on the right, or 2 left, 2 right, respectively).
That doesn't mean, though, that these were actually necessary/used in practice. Clearly, first order particles don't need as many ghost points, and another aside is that the ghost points kinda have double duty, being used for interpolation / current deposition, but also do advance the E&M fields, but for the latter I think one only needs to go [0.. m], IIRC.
I don't know if you thought through in particular the case of 2nd order particles (which aren't really sure to be used again anytime soon/ever). But I think it'd be preferable to keep the #ifdef DEBUG poisoning with NaNs, because that's a straightforward way to make sure one isn't using ghostpoints that haven't been set.
This reverts commit f5765b8.
2 participants
Adds a test to
test_reflective_bcs_integration.cxxthat's the same as the existing one, but along z instead of y. It fails, and the bug is tracked to the current reflection ofpsc_bnd_fields_impl.hxx, which had a double subtraction.This PR also makes field reflection consistently only first-order, i.e., correct to 1 ghost cell, since it's impossible to be consistently second-order with only 2 ghost cells per boundary due to the extra cell that node-aligned quantities require.
That change, plus a removal of some debug stuff, can easily be reverted.