PLIP model
The total potential energy of a configuration made of N atomic positions is first given by \(E_{\textrm{tot}} = \sum_{i = 0}^N E_{i}\) where Ei is the atomic energy. For Ei , we considered a weighted linear combination of descriptors indexed by n:
where :math:` omega_n` is the linear coefficient associated with the descriptor \(X_n^{(i)}\).
Descriptors
The descriptor space is divided into two-body, three-body and N-Body components with following equations:
where \(r_{ij}\) is the distance between atoms \(i\) and \(j\), \(\theta_{ijk}\) is the angle centered around the atom \(i\). The cut-off function is as follows:
Switch function:
Commonly, gaussian descriptors are used with the following equation:
where \(a_{n}\) and \(b_{n}\) are parameters to optimise.
Lasso Model
The set of descriptors are optimised using the Lasso regression method as follows:
where \(\alpha\) is a parameter to control regularization and \(\chi^2_{\textrm{OLS}}\) is ordinary least square objective function