A simple approach to rotationally invariant machine learning of a vector quantity

Machine learning (ML) has become a powerful tool in quantum chemistry, especially in molecular dynamics to predict potential energy surfaces. Fitting scalar values (such as energies) with ML models does not need to account for vectorial properties - the predicted energy is rotationally invariant. Models predicting vector or tensor properties, e.g. dipole moments or polarizability, need to satisfy rotational covariance. It means that various rotations of molecule’s coordinates, reguires correspondingly rotated vectors. Here we target the efficiency of such a goal, developing simple, but accurate and fast approach to machine learn vectorial properties.

Caption photos easily. On the left, a road goes through a tunnel. Middle, leaves artistically fall in a hipster photoshoot. Right, in another hipster photoshoot, a lumberjack grasps a handful of pine needles.

This image can also have a caption. It's like magic.

For more info, check out the paper: (Martinka et al., 2024).

References

2024