If you want to have someone to look into it you should kindly consider to disclose some code, because - as Jens-Peer Kuska used to say - the magic crystal sphere standing behind me did not show it on it's own.
It could be an initialization problem, if one runs first x=x1,x2, than x=x2,x1 without clearing all variables before the second run. It could be a typo preventing the intended symmetry - did you perform successful some formal check of the intended symmetry? It could be a numeric accident, going away with bigger precision set ...
