Glue

[eschmidt42]

Glue

Main Change

• added the Finnis-Sinclair form (Finnis and Sinclair 1984, A simple empirical N-body potential for transition metals glue type empirical potential as a new interaction: src/interaction/glue_fs.jl

Note: The "glue" is a scalar pre-computed for each atom and is a required quantity to compute energies and forces. So simply iterating over atom pairs to compute energy and force contributions is insufficient, without having pre-computed the glue values. Hence, in a few places, the MD code needed to be adjusted to allow for this extra step.

Minor Changes

• added a notebook fs.ipynb to document the usage of the added interaction type
• modified accelerations in src/forces.jl: to enable glue calculation before pair contributions to the forces
• modified potential_energy in src/analysis.jl to enable the use of a potential_energy function for the glue interaction type
• modified src/loggers.jl by adding new loggers which proved useful for debugging: GlueDensityLogger, VelocityLogger, ForceLogger
• modified Simulation in src/types.jl by adding glue_densities as an array where per-atom glue values are stored for the glue potential
• added test/glue.jl as a new test to run to verify the correctness of the predicitons of potential energies, forces using reference values from the Finnis and Sinclair paper linked above as well as testing the behavior of the glue scalar values.

Note: A peculiarity of the Finnis-Sinclair potential is that the potential energy of an atom depends on the scalar glue value of that atom by taking the square root. Hence, cases in which the glue value, for whatever reason, becomes negative, will crash a simulation, since it is not defined in the model how to handle complex potential energy values.

To dos

There are probably more things to be done. The immediately relevant steps would be to:

• add shifted force / potential version of the glue potential
• add parallelism for the glue computation
• add other types of glue potentials, like the Embedded Atom method and
• verify the correctness of the implementation for systems with more than one type of element (the implemented Finnis-Sinclair paper only dealt with single element systems).