Generating training data from trajectory

[jonathan-booth]

I have a few trajectories where, at each frame, I have a text file where each line is an atom and we have values for the atom type, the x,y,z positions, and the x,y,z components of the force vector. Each text file also has the total potential energy. I have trajectories for different sized systems. How do I convert these data into something that Nequip (Allegro really) can work with? I've read the docs but not found an answer sorry if I missed something obvious.

Thanks in advance

[DavidW99]

Dear Jonathan,

You can try to convert your file to extxyz format using ase package. Reference: https://wiki.fysik.dtu.dk/ase/ase/io/io.html. Systems with different cell sizes are OK to be put together as your data set. But, please check if the boundary condition (PBC or not PBC) of your output xyz is correctly set because it can affect the result.

Then in your yaml file, you can use the following to read your data file.

dataset: ase                                                                   
dataset_file_name:  your_data.xyz
ase_args:
  format: extxyz

[DavidW99]

No problem, the example is given in this ase page https://wiki.fysik.dtu.dk/ase/ase/io/formatoptions.html#extxyz.

This is an example format with forces, which is a per-atom value listed in each row:

Lattice="1.0 0.0 0.0 0.0 2.0 0.0 0.0 0.0 3.0" Properties=species:S:1:pos:R:3:forces:R:3 energy=-100 stress="0.0002 0.003 -0.004 0.003 -0.004 -0.005 -0.004 -0.005 -0.006" free_energy=-100 pbc="T T T"
H      14.80702000       9.47114000      14.83362000       0.05630700       0.18844700       0.12262800
C      14.40303000       9.58896000       4.81541000       0.28543000      -0.56192100       0.59003500
Al       0.15995000       9.61979000       9.87331000      -0.15746000       0.51516600       0.14365000

I suggest you to create an instance atom from Atom object following https://wiki.fysik.dtu.dk/ase/ase/atoms.html#list-of-all-methods and do atom.new_array(‘forces’, your_forces_array) to add forces.

[jonathan-booth]

OK thanks. How does the extxyz format work with a trajectory? I have one molecule with different positions and forces at each timestep, and I would like to learn force as a function of atomic positions. So do I have a separate block for each frame in the same extxyz file as in my post above or do I have a separate file for each frame? And sorry about the other question on the Nequip page being a duplicate of this, I posted here and then thought it would be better on the Nequip page and I must have forgotten to delete this one.

[DavidW99]

No worries! Indeed, it will be many blocks of frames. For a trajectory, extxyz contains a stack of single frame and the frames are separated by the number of atoms in the previous frame shown below.

3
Lattice="1.0 0.0 0.0 0.0 2.0 0.0 0.0 0.0 3.0" Properties=species:S:1:pos:R:3:forces:R:3 energy=-100 stress="0.0002 0.003 -0.004 0.003 -0.004 -0.005 -0.004 -0.005 -0.006" free_energy=-100 pbc="T T T"
H      14.80702000       9.47114000      14.83362000       0.05630700       0.18844700       0.12262800
C      14.40303000       9.58896000       4.81541000       0.28543000      -0.56192100       0.59003500
Al       0.15995000       9.61979000       9.87331000      -0.15746000       0.51516600       0.14365000
2
Lattice="1.0 0.0 0.0 0.0 2.0 0.0 0.0 0.0 3.0" Properties=species:S:1:pos:R:3:forces:R:3 energy=-100 stress="0.0002 0.003 -0.004 0.003 -0.004 -0.005 -0.004 -0.005 -0.006" free_energy=-100 pbc="T T T"
H      14.80702000       9.47114000      14.83362000       0.05630700       0.18844700       0.12262800
C      14.40303000       9.58896000       4.81541000       0.28543000      -0.56192100       0.59003500

[KP-303]

Thanks! One final question, it says in the ase docs that an extxyz file must contain the lattice vectors. What if I don't have periodic boundaries? If it's just some atoms in the gas phase in vacuum for instance. Do I just put very large lattice vectors in the file so there will be no periodic images within the cutoff during training?

[DavidW99]

Great question. For gas phase, you can set pbc to be False and leave cell parameter in the ase.Atoms object as default. Then, the feature vector will not take the periodic image of atoms. I think ase.Atom object does not require setting the cell, i.e. lattice vectors. Please refer to https://wiki.fysik.dtu.dk/ase/ase/atoms.html#list-of-all-methods for more info.

However, you should be aware that there is no cell stress for the gas phase, and I think this discussion might be helpful #34.