Because changa allows particles to exist exactly at the box edges, the kdtree will fail and revert to being non-periodic. The effect of this seems to be isolated to calculations that would use smoothing. Checks for edge galaxies indicate no issues with recently calculated profiles for timesteps when this occurs. A possible solution is to wrap the box before generating the kdtree.