I have been running my molten slag simulations with various potentials in LAMMP:S and have been using the Wolf summation. However, after some testing I am wondering whether my settings are reasonable. I had started with values listed in the examples online.
wolf 0.2 8.0
However, my preference is to use a 10A cutoff, and for one of the potentials I have
pair_style hybrid/overlay coul/wolf 0.2 10.0 born 10.0
Is this a reasonable choice for alpha = 0.2 and is there anything else inconsistent.
My simulations appear to work well, but I have done some testing with different cutoffs, as well as 0.1 and 0.3 for alpha. The volume/density of a test slag do show some fairly small differences, but not insignificant. I’m wishing to check whether what I’m doing is appropriate please.
Is there a way to confirm that this is ok please?
The online manual and other info I’ve found online does not seem to indicate the abive.
Much appreciated as always,
Please note that this is not really a LAMMPS problem. LAMMPS will apply whatever you feed it and does not discriminate. If you worry about accuracy, you probably should not be using Wolf summation at all and instead use full long-range Coulomb with a Kspace solver.
Thanks for this Axel. Is there any way to determine what would be a reasonable choice of alpha please? LAMMPS does implement Wolf in its own way, are there some known reasonable default values please? The speed of Wolf will be important to our work. Is there another useful source for this info please?
So are you suggesting I run some calcs with full long-range Coulomb with a Kspace solver, and then compare with what Wolf yields?
Again, what you are asking is off-topic here.
If you want to learn more about the Wolf summation and what reasonable values are, you need to study the published literature and discuss with people that care about your research.
In the speed versus accuracy scale, I personally always come down on the accuracy side (what does it help to get sloppy results fast?), since only in the most extreme cases, you will reach the limit of what can be achieved in terms of performance. Getting access to more computer time is far easier to get these days than to repair one’s reputation after you are known for publishing not so great results. The question of when to use an approximate method or not very much depends on the research and availability (in ReaxFF, as it is implemented in LAMMPS, you have no choice).
But this is just an opinion, not a scientific fact, so you can deal with it any which way you like.