In chemistry, the most common way of representing molecules is to use Cartesian coordinates.
The problem of Cartesian coordinates is that if one rotates or translates a molecule, the coordinates change. However, properties like the total energy of the molecule, the dipole moment or the atomisation energy remain unchanged.
Therefore, if you want to learn the properties of a molecule with a neural network, you want to represent the molecular structure in a way that doesn’t change when you rotate or translate the molecule.
By doing this, you make the training process more efficient because the neural network doesn’t need to learn that many different Cartesian coordinates represent the same structure.
This is where Atom Centred Symmetry Functions (ACSF) come in handy. They are a way of representing a molecular structure which remains the same when the molecule is rotated or translated.
Understanding how they work
Note: in this…
View original post 1,073 more words