Lammps newton on

LAMMPS LAMMPS Beginners
1

Hi,

I am simulating a atomistic polymer system -diblock PSS-PEP.
Below is the input script.
I want to know I am running this with CPUs(no kokkos) and is that ok if I keep newton on in input script.
Will that affect the simulation out put?

log lammps.nve
units real
newton on
#package gpu 2
neigh_modify delay 5 every 1 one 10000
#neigh_modify delay 0 every 1 check yes
boundary p p p
#processors 4 2 2
atom_style full
bond_style harmonic
angle_style harmonic
dihedral_style opls
#improper_style cvff
special_bonds lj 0.0 0.0 0.5
pair_style lj/cut 12.0
#kspace_style pppm 1.0e-4
#kspace_modify mesh 20 20 20
read_data PSS15PEPNa6.data
#read_restart restart_2nvelim remap
reset_timestep 0
#replicate 2 1 1
#write_data PSS9_melt.data
#fix 1 all recenter 0.0 0.0 0.0 units box
#velocity all create 300.0 330083
#fix 1 all nvt temp 300.0 300.0 100
#fix 1 all npt temp 600.0 600.0 100 iso 10.0 10.0 100.0
#group polymer type 1 2 3 4 5 6 7 8
#group one molecule 50 51 94 95 109 110 155
compute 2 polymer gyration
compute 3 center gyration
dielectric 1
fix 1 all nve/limit 0.1

fix 1 all nve

fix 2 all langevin 500.0 500.0 100.0 11434267
#fix 3 polymer freeze
#variable r equal ramp(1200,1200)
#fix 3 polymer indent 100 sphere 0.0 0.0 0.0 140.0 side in units box
#fix 3 all balance 100 xyz 10 1.05 out tmp.balance
#fix 4 all deform 1 x final -150 150 y final -150 150 z final -150 150 units box
#fix 3 all recenter 0.0 0.0 0.0 units box
#fix 2 water shake 0.001 20 0 b 15 a 27
fix 10 all momentum 1 linear 1 1 1
#delete_atoms group one compress yes bond yes mol yes
#displace_atoms nafive move -150 -100 -100
thermo 1000
thermo_style custom step temp press vol density cpu
#thermo_modify flush yes
timestep 1
#reset_timestep 0
#run_style respa 2 4 pair 1 kspace 2
restart 100000 restart_1nvelim restart_2nvelim
dump 1 polymer custom 250000 PSS9_THF6_COM.18a id type xs ys zs ix iy iz
dump 2 all custom 100000 PSS15PEP_nvelim id type xs ys zs ix iy iz mol
run 1000000

2

You subject line says “newton off”, but here you ask about “newton on”. Which one is it?

What does the documentation say?
Have you made a test and observed differences?
And if yes, have you tried to assess the significance of the differences?

Well, you can run on the CPU with KOKKOS and on the GPU without KOKKOS.

3

It was a typo.I corrected the subject line.
Lammps documentation says : Newton on means Newton’s third law on for pairwise and bonded interactions. For most problems, setting Newton’s third law to on means a modest savings in computation at the cost of two times more communication.
I have not tested the systems with newton on.
The computer I use for simulation : kokkos is compiled with Cuda backend and therefore will work only when running lammps with this image on the GPU node.

4

You are just repeating what is written in the doc page. I don’t need to know what is in there, since I have read it myself many times. But you have to read and understand what it says and all of it. Among the parts that you did not quote is actually the statement that matters.

Please please use a little common sense here and understand my questions as questions that you - as an accomplished scientist - should ask yourself in order to gain an understanding of the situation your are facing and to figure out an answer to your concerns.

Again, this is not necessarily true. I have LAMMPS compiled for CUDA and can run on the CPU just fine. I just cannot use KOKKOS without using the GPU. Also, my statement was meant in a more general way meaning that you can compile KOKKOS for threading instead of GPUs and that there is also the GPU package where you can run on GPUs without using KOKKOS. This needs compilation of a different binary, of course, but that is something anybody can do. I mainly wanted to point out that your statement does not hold in general, but only for your specific circumstances.

5

Newton should only affect performance, not numerics. Even with Kokkos CUDA, you can still toggle newton on/off. The default is off but on still works.

6

Thank you