These days I am confused the seed in the velocity command:

velocity mobile create 0.91 87285476 temp new (the units is LJ).

In theory, the seed is random and it will not influence the result. While, I discover that the results are different with the seed changing and I cannot find out the reasons.

There are four two-dimensional density distribution figures at different seeds as follows. These seeds are random. I compute twice at seed1 and the results are almost the same.

Fig 1 Two-dimensional density distribution of Seed1

(velocity mobile create 0.91 87285476 temp new)

Fig 2 Two-dimensional density distribution of Seed2

(velocity mobile create 0.91 82433123 temp new)

Fig 3 Two-dimensional density distribution of Seed3

(velocity mobile create 0.91 52543123 temp new)

Maybe I have misunderstood the seed in the velocity command and I will grateful if you can give me some ideas. Thank you very much.

"In theory, there is no difference between theory and practice. In practice, there is."

But maybe that is not your case. How many simulation steps passed between velocity command and images taken? There is always some time before system "forgets" its initial state and goes to equilibrium, where, indeed, velocity seed should not play any role. See, e.g., http://www.ihed.ras.ru/norman/eLibrary/CompPhysComm_147_678.pdf .

These days I am confused the seed in the velocity command:****

velocity mobile create 0.91 87285476 temp new (the units is LJ).****

In theory, the seed is random and it will not influence the result. While,
I discover that the results are different with the seed changing and I
cannot find out the reasons.****

There are four two-dimensional density distribution figures at different
seeds as follows. These seeds are random. I compute twice at seed1 and the
results are almost the same.

please note that LAMMPS uses a simple, very fast, but rather low quality
pseudo random number generator for the generation of initial velocities.
since MD simulations need to be subjected to equilibration, it is has so
for not been considered important to use a higher quality pseudo random
number generator. LAMMPS contains a second higher-quality RNG (and i have a
third, state-of-the-art RNG class in LAMMPS-ICMS)

if this matters for you, you can try to modify your copy of LAMMPS
accordingly and repeat the tests.