Hi Christina, Steve, et. al.
That (adding epsilon to the right-hand x boundary) did the trick, so presumably the unhappy guy was at x>=L_x to the numerical precision with which I wrote down the box size (using "ascii restarts"). I would have guessed that he would have been mapped to x=zero before the initial MPI broadcast, but I guess this doesn't happen.
This is confusing, since Domain::remap() looks like it should be doing exactly this... should take x=xmax -> x=xmin if there are PBCs in x. Domain::remap() looks like it's being called where I would expect: