[lammps-users] Untraceable reference for fast coulomb solvers

Hello Developers,

In the description for the k-space solvers, I came across the following interesting citation for an article that I couldn’t find:

(Who) Who, Author2, Author3, J of Long Range Solvers, 35, 164-177 (2012).

https://docs.lammps.org/kspace_style.html#who2012

It appears that somebody had intended to add in a reference here but never got around to doing it…but it could very well be a prank too :slightly_smiling_face:

On a more serious note, I am trying to find an appropriate fast Coulomb solver for a fluid of small rigid molecules with long range multipole Coulombic and Lennard Jones interactions to compute its viscosity and thermal conductivity in systems with both periodic and non-periodic boundaries (eventually).

I have found that simulations with lj/long/coul/long pair-style with the ewald/disp summation based kspace-style become very expensive with about 90% of the time spent computing Coulomb interactions.

Any pointers from experienced users on the use of an appropriate fast Coulomb solver in LAMMPS shall be very useful.

Thanks.

Warm regards,
Vaibhav.

The reference you are looking for is (courtesy of the upcoming new LAMMPS paper):

G. Sutmann. ScaFaCoS – a Scalable library of Fast Coulomb Solvers for particle Systems.
In Bajaj, Zavattieri, Koslowski, Siegmund, Proceedings of the Society of Engineering Science 51st Annual Technical Meeting. 2014.

Besides, using long-range coulomb is expensive and can be excessively expensive when run incorrectly.
In general, you have a choice: do you want fast results? or do you want accurate results? there is no implementation that is always both.

Axel.

Thanks very much for pointing out the correct reference.

I am using the following settings for my simulations with rigid ethylene molecules:

units metal
atom_style full
dimension 3
boundary p p p
pair_style lj/long/coul/long long long 14.0 #${lx}
kspace_style ewald/disp 1e-6
pair_modify mix arithmetic
timestep 0.0001

group c2h4 type 1:5
neighbor 0.3 bin
neigh_modify every 1 delay 0 one 10000 page 100000
neigh_modify exclude molecule/intra c2h4

atom_modify first c2h4

comm_modify mode single group c2h4 cutoff 4.0

I am currently surveying different force-fields so am interested in fast results initially but eventually also need them to be accurate as I zero in on a suitable one.

Again, any insights from experience shall be very useful since I need to compute the full pressure tensor and the heat flux for my calculations and not all available Coulomb solvers will support efficient computation of these quantities as per the documentation.

Thanks.

Warm regards,
Vaibhav.

without seeing some logfile output and thus knowing about system size, number of processors, load imbalance and time distribution, it is difficult to make any recommendation. just the input tells only part of the story.

a couple of obvious questions based on the input only:

  • why do you need to do long-range treatment of the LJ r^6 term when you use a 14 \AA cutoff?

  • why do you use “ewald” and not “pppm”?

  • by using “neigh_modify exclude molecule/intra c2h4” you have inconsistent treatment of coulomb interactions between the pair style and the kspace style. why?

  • why such a tiny timestep?

Axel.