Incorrect energy block calculation of matrix A in fit-snap.jl

[emmanuellujan]

The energy block does not seem to represent the SNAP energy block in a multielement system: TiO2. E.g. the number of columns of the energy block (14) is different from the force block (28). Also, the first column of the left and right energy blocks should be equal to the number of atoms of each type in the configuration (8 Ti and 16 O).

Example of how the first column of the energy block is currently calculated:

julia> evaluate_basis(train_sys[1], train_pars[1])'
1×14 adjoint(::Vector{Float64}) with eltype Float64:
 331776.0  6.48762e5  9.44279e5  9.37241e5  1.2034e6  9.09245e5  …  1.14977e6  1.35237e6  1.38483e6  1.3425e6  1.26222e6

On other hand, the force block seems to be ok. E.g.

julia> evaluate_basis_d(train_sys[1], train_pars[1])[1]'
3×28 adjoint(::Matrix{Float64}) with eltype Float64:
 0.0  -0.100626  -0.345134  -0.401752   -0.847065  …  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0
 0.0  -1.31982   -4.53657   -5.16054   -10.9316       0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0
 0.0   1.09027    3.75512    4.32714     9.18019      0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0

Find more information at InteratomicBasisPotentials.jl/LAMMPS/FitSNAP.