CALYPSO-GAP is a machine learning potentials(MLPs) to calculate the properties of multi-species materials.
There are two key methods in machine learning potentials which are descriptors and machine learning model.
In the CALYPSO-GAP, the weighted atom centered symmetry functions(wACSF) is used as descriptors, and wACSF is based upon the original ACSF which was suggested by Parrinello and Behler and describes the atomic environment by a series of radial symmetry functions(RSF) and angular symmetry functions(ASF). To overcome the scaling problem of ACSF, the weight of each atom in ACSF is used to distinguish the contribution from different atoms. This idea is inspired by the work of Nongnuch Artrith,
We use Gaussian process regression(GPR) to map the atomic environment vector to atomic energy, total energy usually is calculated by sum of atomic energy, and atomic force and cell stress is the derivative of total energy, for more information about GPR, please read the paper
GaussianApproximation Potentials: The Accuracy of Quantum Mechanics, without the Electrons
Gaussian Approximation Potentials: A Brief Tutorial introduction

CALYPSO-GAP module was integrated in binary calypos.x, it is available free in charge for non-commerical use by individuals and academic or research institutions. Please register in CALYPSO

These are two ways to use CALYPSO-GAP potentials,
(1) Using CALYPSO-GAP for structure prediction, which is the original intention of developing CALYPSO-GAP.
(2) Combining the python package GAPPY and Atomic Simulation Environment(ASE) to perform calculation, such as phonon calculation, molecular simulations and geometry optimization.