Standard quantities¶
In order for multiple simulation engines to be able use arbitrary metatomic models to compute atomic properties, we need all the models to use the same metadata when handling the same quantity. If your model returns one of the quantity defined in this documentation as output; or use them as input; then it must follow the metadata structure described here.
If you need other quantities as inputs or outputs, you should use custom
quantity with a name containing ::, such as my_code::my_quantity. For
such custom quantity, you are free to use any relevant metadata structure, but
if multiple people are using the same quantity, they are encouraged to come
together, define the metadata schema they need and add a new section to these
pages.
Variants¶
Models can define variants of any quantity, for example to provide the same output at different levels of theory in a single model. For more information on variants, please refer to the corresponding documentation.
Physical quantities¶
The first set of standardized quantities for metatomic models are physical quantities, i.e. quantities with a well-defined physical meaning.
The potential energy associated with a given system configuration. This can be used to run molecular simulations with on machine learning based interatomic potentials.
An ensemble of multiple potential energy predictions, generated when running multiple models simultaneously.
The uncertainty on the potential energies, useful to quantify the confidence of the model.
Forces directly predicted by the model, not derived from the potential energy.
Stress directly predicted by the model, not derived from the potential energy.
Atomic masses
Atomic positions predicted by the model, to be used in ML-driven simulations.
Atomic momenta, i.e. \(m \times \vec v\)
Atomic velocities, i.e. \(\vec p / m\)
Atomic charges, e.g. formal or partial charges on atoms
Heat flux, i.e. the amount of energy transferred per unit time, i.e. \(\sum_i E_i \times \vec v_i\)
The spin multiplicity \((2S + 1)\) of the system, with \(S\) the number of unpaired electrons.
Machine learning quantities¶
The next set of standardized quantities in metatomic models are specific to machine learning and related tools.
Features are numerical vectors representing a given structure or atomic environment in an abstract n-dimensional space.