potential format

[WolframMiller]

I use version 0.6.11 because it is compatible with AtomsBase v0.5

Reading the potential from json file can be done in two formats (new_format=true or new_format=false)

For AtomsCalculators.potential_energy(system,potential) potential is required in new format

whereas

for ACEpotentials.atom_energy(potential, z0) potential is required in old format.

Is planned to make this unique ?
Or should one use AtomsCalculators.potential_energy(system,potential) with system of one atom ?

In order to get the bond energy of a single atom on the surface one needs both configurations with and without atom and
the energy of the atom.
This means one has two the potential in two different styles in the run.

[cortner]

I don't really want to develop 0.6 anymore, but this sounds potentially like a bug that has a very quick fix. Can you produce a self-contained script that shows this problem please?

[tjjarvinen]

The feature you are asking is available with the new format aswell, but not documented. You can call e.g. folowing to access it

julia> using AtomsCalculators, ACEpotentials

julia> fname_ace = joinpath(pkgdir(ACEmd), "data", "TiAl.json")
"/home/teemu/.julia/packages/ACEmd/nuHhp/data/TiAl.json"

julia> pot = ACEpotentials.load_potential(fname_ace; new_format=true)
ACE potential consisting of 3 subpotentials

julia> pot.potentials[1]
OneBody{Float64}(Dict(:Al => -105.5954, :Ti => -1586.0195))

julia> pot_onebody = ACEpotential( [pot.potentials[1]] ) 
ACE potential consisting of 1 subpotentials

julia> Al = isolated_system( [:Al => [0, 0, 1.]u"bohr"] )
FlexibleSystem(Al, periodicity = FFF):
    Atom(Al, [       0,        0,        1]u"a₀")

julia> AtomsCalculators.potential_energy(Al, pot_onebody)
-105.5954 eV

Here is a little bit background on what is going behind the scenes

Version v0.6 has two versions of the potentials the old based JuLIP and newer ACEmd based one that you load when you give the parameter new_format=true. The use of new format is optin, so that old code using JuLIP version work without changes.

ACEmd was made as project to integrate with AtomsBase and it later lead to a development of AtomsCalculators. This was addded on top of the JuLIP based ACEpotentials to v0.6 series. The documentation for ACEpotentials stayed mostly on JuLIP version thus e.g. this is missing from it.

The development then proceeded to the new v0.8 and the old v0.6 was let to be as a legacy version. But as v0.8 is AtomsBase v0.4 and does not work currently you have to use the old v0.6 series to work.

[cortner]

Hm - rereading the original request, I don't actually see an issue with the current implementation. You just have to create three systems.

• sys1 = everything
• sys2 = bulk+surface
• sys3 = just one atom in vacuum

Then the energy difference you are after is

energy(sys1, pot) - energy(sys2,  pot) - energy(sys3, pot)

and you can substitute energy for either AtomsCalculators.potential_energy or for JuLIP.energy.

[cortner]

Arguably the "correct" request would be to provide better convenience functions to carve up a system in different ways. The best place for that would be AtomsBuilder

[WolframMiller]

Indeed, it was more a question on the philosophy.
So the strategy is to set up three systems and apply the same energy function rather than to use internal functions of ACEpotential

I will use this strategy in my KMC tool