Hi,

I am trying to find information about the relative costs of LJ, Morse, and EAM potentials, especially the latter two.

I know that LJ is found to be 2.3x faster than EAM in a paper by Plimpton.

In my own tests, I find that potentials are LJ 1x, Morse 4.4x, and EAM 3.7x fast.

I use a 3D FCC structure of ~10K Copper atoms that are already statically minimized and measure the elapsed time for 1K dynamic iterations. I use the Cu_u3.eam potential.

I suspect that Morse should not be expensive as EAM, or is it?

Hi,

I am trying to find information about the relative costs of LJ, Morse, and

EAM potentials, especially the latter two.

I know that LJ is found to be 2.3x faster than EAM in a paper by Plimpton.

please note, that these kind of comparisons cannot be generalized in this

way.

In my own tests, I find that potentials are LJ 1x, Morse 4.4x, and EAM

3.7x fast.

you mean slower, right?

I use a 3D FCC structure of ~10K Copper atoms that are already statically

minimized and measure the elapsed time for 1K dynamic iterations. I use the

Cu_u3.eam potential.

I suspect that Morse should not be expensive as EAM, or is it?

things are never that easy. EAM is often used with a shorter cutoff

compared to LJ and Morse, so it doesn't not only matter how fast you can

compute the interaction between a single pair of atoms, but also how many

pairs you have to consider. keep in mind that the number of atoms in a

(cutoff) sphere grows O(N**3) with the (cutoff) radius.

there also is the (minor impact) of the choice of compiler and compiler

flags as well as CPU hardware or math library. LJ is more affected by

memory bandwidth, than EAM or Morse, who have more expensive to compute

potential functions.

axel.

axel.

I don’t know what “paper by Plimpton” you are referring to.

You can find benchmark results for various potentials

and tarballs with the files that ran them at this link:

http://lammps.sandia.gov/bench.html#potentials

Steve