Forces not being read into input data

[od-qmul]

Sorry for bothering you with more questions, but I couldn't find the answer in the tutorials.

My input .xyz file includes forces but ACE does not seem to include them in the input data or fit using them:

┌─────────┬──────────┬───────┬──────┬────────┬───────┐
│ Type │ #Configs │ #Envs │ #E │ #F │ #V │
├─────────┼──────────┼───────┼──────┼────────┼───────┤
│ default │ 3908 │ 82653 │ 3908 │ 0 │ 0 │
├─────────┼──────────┼───────┼──────┼────────┼───────┤
│ total │ 3908 │ 82653 │ 3908 │ 0 │ 0 │
│ missing │ 0 │ 0 │ 0 │ 247959 │ 23448 │
└─────────┴──────────┴───────┴──────┴────────┴───────┘

My script reads the data using these lines:

using ACEpotentials
data_file = "./datasets/TiO2trainingset.xyz"
datatrain = read_extxyz(data_file)

DO I have to add any flags to make sure forces are included?

[od-qmul]

It seems that read_extxyz has read in the forces:

julia> datatrain[1].data
Dict{Any, JuLIP.JData{Float64}} with 3 entries:
"forces" => JData{Float64}(Inf, 0.0, SVector{3, Float64}[[0.392327, -0.931399, 1.00845], [-0.392327, 0.931399, -1.00845], [0.3921…
"stress" => JData{Float64}(Inf, 0.0, [0.0 0.0 0.0; 0.0 0.0 0.0; 0.0 0.0 0.0])
"energy" => JData{Float64}(Inf, 0.0, -19905.3)

[cortner]

probably the default force key is different, I'll check and get back to you

[cortner]

solver = BLR()  # (for example)
data_keys = (energy_key = "energy", force_key = "forces", virial_key = "stress")
acefit!(model, datatrain; solver=solver, data_keys...)

the data_keys tuple specifies under which key to find the training data. See e.g. in [this tutorial].

Here, I'm naively assuming that stress = virial, which is technically false. So if this is the actual stress, then you need to first go through all training data, rescale it, write it into a new field "virial" and adapt the script.

[cortner]

... and then probably file an issue that we provide a convenience function to do this automatically.