Dear Ray,
Thanks a lot for your reply, sorry for disturbing you again. There was a mistake in unit, you said right, now the stress along x is nearly same with some exp results. But this stress variable Pxx * V = Wxx + Sum[mi*(vxi-ux)*(vxi-ux)],where Wxx = Sum(Force_xx of i) / (LyLz) is a global variable, i.e. it is computed as a global vector of a region, can I make it a “atom style” variable, so that I can do ave/spatial to see the spatial gradient of Pxx with distance ( like as “stress/atom” within lammps), currently I can study evolution of Pxx with time ( by ave/time ).