pair_style on CPU and GPU

LAMMPS LAMMPS Mailing List Mirror
1

Dear Lammps users,
What is difference between “pair_style lj/cut/coul/long/gpu” and “lj/class2/coul/long”??
I want to use lammps on GPU and my force fied is PCFF(Class2). The closest pair_style I found is lj/cut/coul/long/gpu, but the result of my simulation on CPU and on GPU are different. Is there anyway that I get equivalent as “lj/class2/coul/long” on GPU?

Many thanks
Fatima

2

Dear Lammps users,
What is difference between "pair_style lj/cut/coul/long/gpu" and
"lj/class2/coul/long"??

one runs on the GPU, the other doesn't. one implements a 12-6 lennard
jones, the other a 9-6 lennard jones. this is all well documented in the manual.

I want to use lammps on GPU and my force fied is PCFF(Class2). The closest
pair_style I found is lj/cut/coul/long/gpu, but the result of my simulation
on CPU and on GPU are different. Is there anyway that I get equivalent as

of _course_ they are. you are using different potentials. why should
they be the same?

axel.

3

Thanks Axel for your response. I see only “lj96/cut/gpu” is available for 9-6 potential. I need to include coul and long-range interactions as well. Is there any pair_style I can use for that on GPU?
Thanks,
Fatima

4

Thanks Axel for your response. I see only "lj96/cut/gpu" is available for
9-6 potential. I need to include coul and long-range interactions as well.
Is there any pair_style I can use for that on GPU?

your careless treatment of the details of the functional form
of the potentials is scaring me. there is a 9-6 lj in the cg/cmm/coul/long/gpu,
but the compass formulation is different, and i don't see right away, how
you can translate the class2 parameters into this or the other supported
functional forms.

if you _have_ to have lj/class2/coul/long/gpu you will have
to write it or make a convincing case to somebody else
to write it for you.

axel.