compute MSD on a triclinic box

Dear LAMMPS users,

When I tried to compute MSD on a triclinic box, even though the triclinic factors very small, say: triclinic box = (0 0 0) to (81.5697 81.5696 81.5696) with tilt (0.0407848 0 0)

I found the curve of MSD_x vs time is reasonable: first increase rough linearly and then flatten, MSD_x = ~0.002 A^2
But the MSD_y and MSD_z are crazy large. Even when t = 0, MSD_y = 1108.24 A^2 and MSD_z = 1111, and their variation with time is smaller than 0.03 A^2. These are obviously unreasonable, could you tell me the possible reasons? Thanks in advance.

the commands for msd are below: (I’m calculating vibrational MSD, so set average yes)
fix 1 all nve
compute MSD all msd com yes average yes
fix out_msd all ave/time 1 1 1 c_MSD[1] c_MSD[2] c_MSD[3] c_MSD[4] file msd.dat
run 10000


Can you post a simple-as-possible input script which illustrates
the problem. Are you changing the triclinic box over time,
e.g shearing the system? As you time monitor the MSD components
do they change dramatically at points in time, e.g. when the triclinic
box is remapped to a different shape?