LAMMPS: Inquiry on Random Seed and Data Reproduction"

Dear Dr. Axel Kohlmeyer,

Hello Dr. Kohlmeyer. Thank you for your e-mail.

I was curious if my understanding is correct.

If my input script contains a command line such as “velocity all create 1.0 1483274” and run this script many times, the random seed will ensure that I will reproduce the exact same results when I run the same script multiple times.

However, if my input script has “pair_style dpd 1.0 1.0 1224589” or any other command such as “fix langevin” that has a thermal input argument, there is no guarantee that I will reproduce the exact same results for every same run due to the inevitable “thermal noise.”

Furthermore, if I have dpd pairwise command where the pairwise coefficient frequently changes for every 50 timesteps like the example below, the random seed will not be able to reproduce same results.

pair_style dpd 1.0 1.0 122347

pair_coeff * * 25 4.5 1.0

run 1000

pair_coeff * * 50.0 4.5 1.0

velocity all create 1.0 15086120

run 50

pair_coeff * * 100.0 4.5 1.0

velocity all create 1.0 15786120

run 50

pair_coeff * * 150.0 4.5 1.0

velocity all create 1.0 15486120

run 50

Overall, I was curious if you suggest that these slight variations in data reproduction should not be an issue since the deviation is marginal.

Thank you.

Sincerely,

Masato Koizumi

Dear Dr. Axel Kohlmeyer,

Hello Dr. Kohlmeyer. Thank you for your e-mail.

I was curious if my understanding is correct.

If my input script contains a command line such as "velocity all create
1.0 1483274" and run this script many times, the random seed will ensure
that I will reproduce the exact same results when I run the same script
multiple times.

​why don't you just read the documentation? start here: ​

However, if my input script has "pair_style dpd 1.0 1.0 1224589" or any
other command such as "fix langevin" that has a thermal input argument,
there is no guarantee that I will reproduce the exact same results for
every same run due to the inevitable "thermal noise."

Furthermore, if I have dpd pairwise command where the pairwise coefficient
frequently changes for every 50 timesteps like the example below, the
random seed will not be able to reproduce same results.

pair_style dpd 1.0 1.0 122347

pair_coeff * * 25 4.5 1.0

run 1000

pair_coeff * * 50.0 4.5 1.0

velocity all create 1.0 15086120

run 50

pair_coeff * * 100.0 4.5 1.0

velocity all create 1.0 15786120

run 50

pair_coeff * * 150.0 4.5 1.0

velocity all create 1.0 15486120

run 50

Overall, I was curious if you suggest that these slight variations in data
reproduction should not be an issue since the deviation is marginal.

​go and pick up a text book on statistical thermodynamics​ and learn why.

as i have been pointing out before. this *exact* same issue has been
discussed multiple times on this mailing list. in some cases at great
length.
go and study those discussions properly and your questions will be answered.

axel.

Thank you.

Sincerely,

Masato Koizumi

​--​

Dr. Axel Kohlmeyer [email protected] Homepage of Axel Kohlmeyer
College of Science & Technology, Temple University, Philadelphia PA, USA
International Centre for Theoretical Physics, Trieste. Italy.