bin size for hybrid pair style

Dear LAMMPS developers,

I have a question about the skin of the Stillinger-Weber potential:
I have a system with hybrid pair style consisting of lj/cut, yukawa, and sw, with the cutoff of 14, 10, and 4.3 Angstroms, respectively. (the cutoff of the sw potential is defined by sigma*a). The cutoff of the lj/cut is much larger than that of the sw potential.

I notice that for the “neighbor skin bin” style, the bin size is set to 1/2 of the largest cutoff distance between any pair of atom types and a single set of bins is defined to search over for all atom types. So my question is: how is the bin size set for different pair styles if I use “neighbor skin bin” style? Do I need to use “neighbor skin multi” to get higher efficiency?

Thank you so much!

Best regards,
Jibao

Dear LAMMPS developers,

I have a question about the skin of the Stillinger-Weber potential:
I have a system with hybrid pair style consisting of lj/cut, yukawa, and
sw, with the cutoff of 14, 10, and 4.3 Angstroms, respectively. (the cutoff
of the sw potential is defined by sigma*a). The cutoff of the lj/cut is
much larger than that of the sw potential.

I notice that for the "neighbor skin bin" style, the bin size is set to
1/2 of the largest cutoff distance between any pair of atom types and a
single set of bins is defined to search over for all atom types. So my
question is: how is the bin size set for different pair styles if I use
"neighbor skin bin" style?

​from the documentation of the neighbor command:

For style multi the bin size is set to 1/2 of the shortest cutoff distance
and multiple sets of bins are defined to search over for different atom
types.

Do I need to use "neighbor skin multi" to get higher efficiency?

​why don't you just make a test and see which setting is more efficient?

axel.​

Dear Axel,

Thank you. Is the neighbor list for the three-body interactions (as in the sw potential) handled in the same way as in the pair-wise interactions (as in the lj/cut potential)?

“For style multi the bin size is set to 1/2 of the shortest cutoff distance and multiple sets of bins are defined to search over for different atom types.” I don’t understand how the multiple sets of bins are defined. Are they defined atomatically by lammps? Could you please give me a brief explanation on that?

I appreciate your time and help!

Thanks,
Jibao

Dear Axel,

Thank you. Is the neighbor list for the three-body interactions (as in the
sw potential) handled in the same way as in the pair-wise interactions (as
in the lj/cut potential)?

​why do you worry about this?​

"For style multi the bin size is set to 1/2 of the shortest cutoff
distance and multiple sets of bins are defined to search over for different
atom types." I don't understand how the multiple sets of bins are defined.
Are they defined atomatically by lammps? Could you please give me a brief
explanation on that?

​if you want to know things at that level, please read the source code.

axel.​

Dear Axel,

Actually, I’m not intending to know things at that level exactly. Because the manual for neither the pair_style sw potential nor the neighbor mention how the neighbor list of the sw potential is handled, I just want to make sure that I’m doing the correct thing (using “style multi bin size” for the hybrid pair style containing sw, lj/cut, and yukawa, with the cutoff of sw much smaller than that of the lj/cut (4.3 vs. 14)). If you don’t think there is any problem of doing this, I would just run the simulation instead of going into the details of the source code.

Thanks,
Jibao

Dear Axel,

Actually, I'm not intending to know things at that level exactly. Because
the manual for neither the pair_style sw potential nor the neighbor mention
how the neighbor list of the sw potential is handled, I just want to make
sure that I'm doing the correct thing (using "style multi bin size" for
the hybrid pair style containing sw, lj/cut, and yukawa,

​neither does the manual say how neighbor lists for any *other* potentials
are handled. ​there is a long list of neighbor list construction variants
that depend on all kinds of settings.

with the cutoff of sw much smaller than that of the lj/cut (4.3 vs. 14)).
If you don't think there is any problem of doing this, I would just run the
simulation instead of going into the details of the source code.

​like i already mentioned, the only way to make certain is to run tests.
you need to verify your model anyway, regardless of the choice of neighbor
list construction.
you could with the same justification ask about how "newton on/off" or
"run_style verlet" vs "run_style respa" affects neighbor lists. or using
gpu, kokkos, and so on.​
​and what about the time step? the number of processors? the list is
endless.​

​just do it,​ and see for yourself.

​axel.

Hi Axel,

I will do the test. Thank you. I appreciate your comments.

Jibao

I suggest you don’t use neihgbor multi. Try it

first with regular neighboring. I don’t think it

will buy you anything for a hybrid pair style.

Steve

Hi Steve,

Thank you for your advice. I will try that.

Regards,
Jibao

Also, you can still play with the bin size for

neigh style bin.

Steve