[FLASH-USERS] Nonconvergence in subroutine rieman!

Hi all,

    I met an annoying problem in the rieman.F90. It takes me a few days
to pin down but without any progress.
I post here to hope someone may meet the same problem or have some good
ideas to get ride of it.

In the attachment (a.eps) I listed the value around the annoying points
in two consecutive steps.
The values in the first column looks normal. But the values of one zone
for next step (right column) become
totally unbelievable, which are marked by red rectangle.
This happened at the +Y boundary (the simulation box contains 64^3 zones).
Are there anyway to get rid of the problem with less effort?
Thanks very much!

Best wishes,
Shikui